69 builds · one impossible brief

FIELD INDEX
2026 / 01

Kill My SaaS
Field Notes

Every competitor-built conference product, captured in the wild. Browse the interfaces, declared AI tools, technology signals, and build economics.

Explore the builds↓

69of 69
showing

Search index

Visit ↗

OpenSession

opensession-web.vercel.app

Open demo

VercelNext.js

Agents: Claude Code · Codex
Models: Claude Fable · Claude Opus · Claude Sonnet/Haiku
Spend / usage: $500 reported

Process & submission notes +

Build process

I used herdr running on my hetzner vps, fable 5 to run for the first 24h then switched to Opus that was using codex as sub agents.

Additions & omissions

I've created a migration file here: [link removed] The style/UI especially for public facing pages is left neutral and clean: [link removed] So if you want i can style it however you like :)

Visit ↗

ProgramLoom

programloom.com

Open demo

Cloudflare

Agent: Codex
Model: GPT 5.6 Sol
Spend / usage: $4 reported

Process & submission notes +

Build process

Started on Saturday with an info-dump of the criteria, as well as every piece of info about SessionBoard that I could find! Specifically optimized for a friendly experience compared to what you already use, then put my own twist on it as the build progressed.

Additions & omissions

Tried to make it as judge-friendly as possible, in addition to the most important part which is user-friendliness specifically for less-technical people (which you stated composes most of the team).

@maddiedreese

Visit ↗

Jooevents Demo

jooevents-demo.joocorp.com

Open demo

SvelteKitCloudflare

Agents: Claude Code · Codex
Models: Claude Fable · Claude Opus · GPT 5.6 Sol
Spend / usage: $999 reported

Process & submission notes +

Build process

A lot of steering. I was very involved the whole 3 days, and I'm glad I could as well these days. I did start late though on Monday. I did spend a day and a half planning first before I set them loose on the coding. I also did it based on how I would use it as a non-event-organizer - so you'll see the way it was laid out makes it easier while still maintaining the terminology lightly like CFP.

Additions & omissions

I think it's best to try it. It should be pretty self-explanatory I hope! And reliefs you of all the busywork and it should be intuitive, with tooling to help you do your job like reviewing, collecting CFP better.…

@thebowlboy

↻ Consolidated resubmission

Visit ↗

Greenroom

greenroom.usespaces.dev

Open demo

Next.jsCloudflare

Agents: Claude Code · Codex
Models: Claude Fable · Claude Opus · GPT 5.6 Sol
Spend / usage: $75 reported

Process & submission notes +

Build process

I started by treating this as a product-design problem, not just a coding exercise. I used fable to turn the competition brief, organizer clarifications, Sessionboard documentation, and reference walkthroughs into a detailed specification. I had the agent separate the essential conference workflow from optional features, document important decisions and tradeoffs, and research uncertain areas before committing to an implementation.…

Additions & omissions

Additions Airtable sync: production DB data is sync'ed to specified Airtable base so Airtable automations can run when data lands there API and MCP: production deployment has API and MCP using token and Oauth so you can use agents to run the show. CRM: one searchable directory with tags, saved segments, event history, bulk email, and a sourcing pipeline

@yogeshkd

Visit ↗

OpenBoard

openboard.alexlazar.dev

Sign-up

Cloudflare

Agents: Claude Code · OpenCode
Models: Claude Fable · Claude Opus · GPT 5.6 Sol
Spend / usage: $100 reported

Process & submission notes +

Build process

I use the [person removed] skills. I actually spent two full days just grilling and working through what I actually want to build and how I want the system or architecture to look. After I had a good spec down and GitHub issues for it to be built, I just basically drove my background agents through the tasks, one by one, sometimes in parallel, to get them done.…

Additions & omissions

I consciously skipped a lot, as you will probably notice, and the reason is that I wanted to make sure that what I do have is very reliable and polished. My thesis is that you would rather have less, but it really works well, especially given we only had a few days to do this.

Visit ↗

Open Session

session.drawset.com

Sign-up

ReactCloudflare

Agents: Claude Code · Codex
Models: Claude Opus · GPT 5.6 Sol
Spend / usage: $300 reported

Process & submission notes +

Build process

I first tried to understand the Session Board because I had never used it before, so it was new to me, and used my AI tools to understand each feature we want to build. Initially felt from your demo the UI can be redesigned, and I also decided that we can retain the core structure of it, as most of the time end users are going to be non-technical, so it should not be an issue for them to switch to Open Session without learning…

Additions & omissions

Added basic Speaker CRM features and API access to event data. We never store passwords or integrate any external auth provider — that's a deliberate cost/complexity tradeoff, since email-based sign-in can reuse whatever mail provider you're already using for the app's other emails. We also have clear docs for hosting this either on your own VM or on Cloudflare.

Visit ↗

GatherPulse

gatherpulse.xyz

Sign-up

Stack not detected

Agents: Claude Code · Codex · Antigravity
Models: Claude Fable · Claude Opus · GPT 5.6 Sol
Spend / usage: $230 reported

Process & submission notes +

Build process

See gist for writeup [link removed]

Additions & omissions

To be totally honest, I focused a lot more on the process than the end result on this one. I didn't realize we might get reimbersed so really just used it as a learning experience and a project to get my process more refined. The UX unfortunately has a lot of roughness because it started are a rough clone focused mostly on functionality parity.…

Visit ↗

bodo

bodo.nsawinyh.workers.dev

Open demo

Cloudflare WorkersNext.jsCloudflare

Agents: Claude Code · Codex
Models: Claude Fable · Claude Opus · GPT 5.6 Sol
Spend / usage: $400 reported

Process & submission notes +

Build process

I started from the incumbent rather than from a blank page: 42 screenshots of Sessionboard were transcribed into per-screen parity docs covering layout, controls, labels and defaults, so the target was a concrete artifact instead of my own guess at how a form builder should work.…

Additions & omissions

Here is a doc on what was built on top of original doc: [link removed]

@sneg55@sawinyh

Visit ↗

Opensession

opensession.opensession.workers.dev

Open demo

Cloudflare WorkersReactCloudflareCloudflare D1

Agents: Claude Code · Codex
Models: Claude Fable · Claude Opus · Claude Sonnet/Haiku
Spend / usage: $141.4 reported

Process & submission notes +

Build process

a) Read the eval kit before writing any code and treated it as the spec. b) Day 1: research, execution plan, and a starter repo with the full 40 plus table schema, seeded demo data, and build rules (CLAUDE.md, DESIGN.md). c) Days 2 to 3: five phased Claude Code prompts (auth and events, CFP and forms, review and decisions, portal and content, widgets, email, Airtable, API, CRM), deploying to Cloudflare Workers and D1 after every phase…

Additions & omissions

Notable Additions or Omissions: a) Additions - email is real, not simulated (acceptance and schedule emails deliver with calendar invites attached, every send logged in app); automated task deadline reminders on a cron with 24 hour dedupe; two way Airtable sync so a non technical team can edit in a spreadsheet and it flows back; a public REST API with docs at /docs/api; zero JS…

Visit ↗

SuperStage

superstage.alizaidkhoso.workers.dev

Sign-up

Cloudflare WorkersNext.jsCloudflare

Agents: OpenCode · Cursor · Antigravity
Model: Gemini
Spend / usage: $100 reported

Process & submission notes +

Build process

I started by understanding the requirement and its workflow. I took screenshots of the session board from the reference video and gave them to the coding agent to understand the UI and plan the implementation. then broke the platform into smaller flows and built them with Cursor and OpenCode. I focused on getting the core workflows working end-to-end first, then tested, fixed bugs, and polished the UI before deployment

Additions & omissions

Added the core end-to-end speaker workflow with CFPs, reviews, onboarding tasks, scheduling, emails, calendar invites, and public event pages. I intentionally skipped multiple reviewer invites to stay focused on the core features and hackathon scope.

@heyalizaid

Visit ↗

Greenroom Events

killmysaas.greenroom-events.workers.dev

Demo login

Cloudflare WorkersCloudflare

Agents: Claude Code · Codex
Models: Claude Fable · Claude Opus · GPT 5.6 Sol
Spend / usage: $500 reported

Process & submission notes +

Build process

Spec-first: cloned the official killmysaas-evals harness on day one and treated specs/*.yaml as the requirements doc. The plan (docs/PLAN.md in the repo) allocates effort by rubric weight and item type. The eval README says exists/crud items rarely discriminate and clones fail on rule/scoping/handoff — so the build prioritized server-enforced rules (CFP open/close window, role scoping that returns 404s rather than empty lists,…

Additions & omissions

Additions: org-level speaker CRM (extra credit) with a sourcing pipeline and real returning-speaker stats; five embed formats (JS loader, JSON, XML, iCal) plus a public event site; per-attendee personal schedule with .ics export; blind review with weighted scorecards; an Outbox screen showing every email the system generated.…

@kh_firdavs

Visit ↗

ChartStead

chartstead.com

Open demo

AstroCloudflare

Agents: Codex · OpenCode · Pi / Oh-my-pi
Models: GPT 5.6 Sol · GPT 5.6 Terra/Luna · Grok
Spend / usage: $300 reported

Process & submission notes +

Build process

Well, I spent a long time planning it out. Looking back, I probably spent too much time planning and really focusing on general pain points and best practices instead of the actual requirements that were listed. Then I started implementing, doing lots of research and planning with intelligent models and design with Codex and soul, and implementation and coding with mostly Grok. I used Grok a lot, mostly Grok to build the whole thing.…

Additions & omissions

First of all, just know there's about a day's work left to really get this to where I want it to be. But I could easily spend another week on it, thoroughly implementing all the features and polishing it that I want. But the main thing that I put in here that set me apart from everyone else is a feature I called Course Check.…

Visit ↗

Greenroom

greenroom-hq.com

Open demo

VercelNext.jsPostgres

Agents: Claude Code · Codex · Pi / Oh-my-pi
Models: Claude Fable · Claude Opus · GPT 5.6 Sol
Spend / usage: $449 reported

Process & submission notes +

Build process

I ran two independent agent teams in parallel and made them adversarial to each other. Each track worked in its own git worktree against its own database. Neither could merge its own work: the other track re-ran the full gate at the exact commit SHA and had real authority to refuse.…

Additions & omissions

WORTH YOUR ATTENTION: - Acceptance is one transaction. Accepting a proposal creates the confirmed session and assigns every speaker's onboarding tasks together, or does neither. No half-accepted talks. - Scheduling refuses conflicts inside the write, using a Postgres advisory lock, with an executable concurrency proof that fails rather than skips when you opt into it.…

@jeremyjem

Visit ↗

OpenSpeaker evaluator access

openspeaker-kms.premhiru.chatgpt.site

Open demo

Cloudflare

Agent: Codex
Model: GPT 5.6 Sol
Spend / usage: $60 reported

Process & submission notes +

Build process

I started from the project brief and the YouTube video to generate a project plan. Once the LLM eval came live, I asked codex to fit that criteria

Visit ↗

Greenroom

greenroom.kwan765.workers.dev

Open demo

Cloudflare WorkersReactCloudflareCloudflare D1

Initial inspection

⚠ Expected an open demo but the submitted route landed on sign-in.

Agents: Claude Code · OpenCode
Models: Claude Fable · Any China open model
Spend / usage: $188.51 reported

Process & submission notes +

Build process

Started from the brief's requirements and mapped each to a concrete screen plus a Drizzle data model (events → forms → submissions → evaluation plans → sessions → agenda → tasks). Built feature-first in dependency order: schema + rich demo seed, then a form builder with per-field conditional logic, close dates and submission limits; then the public CFP wizard; then the speaker portal (profile, headshot/slides, file requests, R2 uploads)…

Additions & omissions

Added beyond the brief: public read-only REST API (/api/v1), /llms.txt for AI agents, embeddable self-contained HTML feeds (agenda/speakers/sessions) with copy-code, XLSX export + CSV session import (Accelevents import path), per-speaker outstanding-task lists, .ics invites with a full Outbox email audit log, and an Airtable sync you can point at your own Base.…

@kirin765

Visit ↗

Sessionbored

sessionbored.com

Demo login

ReactTailwind CSSSQLite

Agents: Claude Code · Codex
Models: Claude Fable · GPT 5.6 Sol
Spend / usage: $350 reported

Process & submission notes +

Build process

Requirements first: I distilled the video walkthrough and the brief into docs the agents could work from, ran four research spikes (Sessionboard's API, Cloudflare, Airtable, Accelevents), then locked a 29-table data model and a written milestone plan before any feature code. Build: Claude Code (Fable) ran as the orchestrator.…

Additions & omissions

There’s an assistant on every page. You tell it what you want in normal words and it does the actual work — moves submissions, sends the emails, edits the form, puts a talk on the schedule, builds a dashboard, exports a spreadsheet. If something would go wrong, like a room clash, it stops and tells you instead of doing it.…

@thedevakrishna

Visit ↗

Conference Program Manager

app.namos-sessions.xyz

Sign-up

ReactCloudflare

Initial inspection

⚠ App is served over plain HTTP rather than HTTPS.

Agents: Claude Code · Codex
Models: Claude Opus · Claude Sonnet/Haiku · GPT 5.6 Sol
Spend / usage: $342 reported

Process & submission notes +

Build process

I had my whole team — me and my AI agents — actually read through all of Kill My SaaS's documentation before touching anything, instead of skimming it. I ran the build through my own planning system, Namos Plan: figure out what's already there, write down exactly what needs to happen and why, turn that into a real tracked task, then build it — so nothing got built without being thought through first.…

Additions & omissions

This is free and open source, but it's built to the same bar as the enterprise tools that run over $40k a year — not a stripped-down version of one. It's agent-native and voice-first: you can talk to it instead of clicking through forms, and it handles the real workflow end to end — CFP intake, review and scoring, scheduling, and speaker emails — with real templates already…

Visit ↗

Conferencer

conferencer.cc

Manual access

Next.jsReactPostgresSQLitePython

Agent: Cursor
Models: Claude Opus · GPT 5.6 Sol · Grok
Spend / usage: $750 reported

Process & submission notes +

Build process

my workflow: 1. parse Google doc (incl description + screenshots) 2. perform reconnaissance on SessionBoard (full crawl of their site + learning pages via Firecrawl) 3. generate plan using Codex GPT 5.6 Sol High 4. implement using Opus 5 via Cursor (GRIND mode engaged) 5. run for 4.5 hours straight❗️ 6. iterate + feedback via Cursor CLI (~10 turns)

Additions & omissions

- Creating a conference/event is via CLI / DB, not the UI. We preload a simple default here: [link removed]

Visit ↗

Lectern

lectern.lectern-go7.workers.dev

Demo login

Cloudflare WorkersCloudflare D1Cloudflare

Agents: Claude Code · Codex
Models: Claude Fable · Claude Opus · Claude Sonnet/Haiku
Spend / usage: $509.59 reported

Process & submission notes +

Build process

Read the brief, built the full program pipeline on one Cloudflare Worker + D1 + R2: CFP with conditional logic → committee decisions with human-reviewed drafted emails → conflict-aware agenda → speaker portal → published schedule and embeds. Then made your eval kit the boss: ran sbek against our live deployment and went 37% → 93.5% at 100% rubric coverage in a day, committing every verdict, the counter-evidence, and the gap ledger to…

Additions & omissions

Additions worth your attention: no accounts required anywhere (capability links + "email me my links" recovery; optional submitter accounts if you want a dashboard) — a solo organizer can escalate to a committee without user management, and committee notes stack per reviewer name. Every email is receipted in the outbox; real Resend delivery is proven with provider message ids.…

Visit ↗

Greenroom

greenroomplanner.com

Sign-up

Next.jsCloudflare

Agent: Claude Code
Model: Claude Fable
Spend / usage: $75 reported

Process & submission notes +

Build process

I first just downloaded everything: - the Google Doc - the Twitter threads - I scrolled through Discord and saved any useful back-and-forth - I took the YouTube video and made a transcript - I took PNG screenshots of the key parts of the workflow - I used speech-to-text to dump a lot of information about the competition and what we wanted to do and all of that - I put it all in one directory - I started up a fable session using the…

Visit ↗

OpenSession

opensession.dev

Sign-up

Cloudflare WorkersCloudflare D1ReactSQLiteCloudflare

Agent: OpenCode
Models: Claude Fable · Claude Opus · GPT 5.6 Sol
Spend / usage: $400 reported

Process & submission notes +

Build process

Spent first 3 sessions researching the problem, reading all requirements and Google doc, trying out sessionize.com to get a sense of the problem and the product. Also reading all API docs and understanding all entities in the system of record. Then spent the first session using Fable to architect the database schema, using Prisma syntax as presentation format.…

Additions & omissions

I didn't implement an Airtable integration because the requirements were not very clear, but the platform has an API you can integrate yourself directly using an agent. It is flexible enough to do basically everything the platform can do in the browser.…

@__morse

Visit ↗

Unsession

unsession.dev

Open demo

PostgresCloudflare

Agents: Claude Code · Codex
Models: Claude Fable · Claude Opus · GPT 5.6 Sol
Spend / usage: $200 reported

Process & submission notes +

Build process

I spent the majority of my time talking to Fable within Claude Work and Claude Design. I’ve never organized a conference, so I spent a lot of time learning and “feeling out” the core user flows. I loosely applied a jobs-to-be-done perspective, and trimmed as much friction as I could find. Once I landed on a solid prototype, I took it into Claude Code and implemented with Fable. V1 was nice, but it had loads of UX issues.…

Additions & omissions

I added an MCP so that you can interact with unsession through your agents. I also included an API as requested. I chose not to include the AI Agenda Builder because the heuristics for a "good" agenda builder weren't clear to me beyond obvious things like room/speaker/time collisions. I didn't see what an AI would provide that deterministic rules couldn't.…

Visit ↗

Openboard

openboard.events

Sign-up

Next.jsCloudflare

Agents: Claude Code · Codex
Models: Claude Fable · Claude Opus · Claude Sonnet/Haiku
Spend / usage: $400 reported

Process & submission notes +

Build process

So the above requires a number, but I used my Claude 200 max + ChatGPT 200 (+ 3x resets, thanks [person removed]). Process was generally similar as what I normally do. Detailed plan with parallelizable tasks first, implementation, then lots of rounds of bugfixing. I think the main differences here were using a lot of YOLO mode to avoid the slowdowns of even auto mode and the Sonnet-heavy dynamic workflows.

Additions & omissions

Spent most of the last two days on bugfixing and cleanup (800 commits and counting!) I think one of the big things is to try to put the code base in a maintainable, well architected, state, but not sure how you'd judge that. Also there are some easter eggs.

@yisding

Visit ↗

opensesh

app.opensesh.io

Open demo

CloudflareCloudflare WorkersCloudflare D1Postgres

Agents: Claude Code · Codex
Models: Claude Fable · GPT 5.6 Sol
Spend / usage: $500 reported

Process & submission notes +

Build process

I started by handing codex the google doc, youtube video, a discord thread, and asked it to get all the context it can. download stuff from SessionBoard docs, videos, transcripts, take screenshots. create a comprehensive PRD of what it is that we are trying to build. Then I spent some time trying to understand the product deeply. Once I got it, I start off. I like to make fable do architecture, specs, review, taste.…

Additions & omissions

I love linear. I want software to always feel like that. load instantly, never lose my position in a table, never be scared to click something and then needing to find my way back. thats where the focus went. every list keeps your scroll and your filters, details open next to the table instead of on top of it, every state is a URL, back always works.…

@earth_ish@d4mr

Visit ↗

untitledconference

untitledconference.com

Demo login

SvelteKitCloudflare

Initial inspection

⚠ Submitter described this deployment as unfinished, unavailable, or fragile.

Agents: Claude Code · Codex · Pi / Oh-my-pi
Models: Claude Fable · Claude Opus · GPT 5.6 Sol
Spend / usage: $600 reported

Process & submission notes +

Build process

Four days, one human, about ten agents, all coordinated in one Buzz channel with a mix of harnesses and models. First: requirements engineering and UX design, before any code. I took every input I could get — your video, the Google Doc, and the website of the tool you use today — and pulled it together into jobs-to-be-done and user journeys. Your eval harness went in as a source too, as a floor rather than a target.…

Additions & omissions

Worth your attention: the MCP server is the part I would test first. I think it is simply not optional any more that a tool works with your own agents and provides a UI. Everything that is normally a lot of clicking — triaging submissions, checking an agenda, making accept/reject decisions — an agent should be able to do for you; the visual part I still want in the UI.…

@getzenai

Visit ↗

Program Os

program-os.isaac-cw-zhang.workers.dev

Demo login

Cloudflare WorkersNext.jsCloudflare

Agents: Claude Code · Codex · Cursor
Models: Claude Opus · GPT 5.6 Sol · GPT 5.6 Terra/Luna
Spend / usage: $200 reported

Process & submission notes +

Build process

I used Sol on XHigh with lots of Luna subagents at the start to build as much "skeleton" as possible. Even if the content wasn't very good, the bones of the application needed to be built and this got it done. Then, when it finished I started using Opus 5 medium to go through every page.…

Additions & omissions

The automated emails are templated and were worked on super hard by me, please check them out I spent most of my time perfecting the flow of agendas, forms, and evaluations, since in my opinion that's the core product. So spending some time on there would be super good! CRM was added albeit hastily, so it might not be super amazing.…

Visit ↗

SpeakerWeave

speakerweave.com

Open demo

Tailwind CSSCloudflare Workers

Agents: Claude Code · Codex
Models: Claude Fable · Claude Opus · GPT 5.6 Sol
Spend / usage: $999 reported

Process & submission notes +

Build process

At a high level: 1. Research - got the agents to analyze 1. Your tweet + doc 2. Sessionboard.com, take screenshots of its UI, analyze user reviews, etc 3. Analyze any conversation in discord 2. Plan with other agents and myself 3. V1 took about 8 hours: fable managing codex agents 4. Did eval and got ~60% - started hill climb to 100% for V2 5.…

Additions & omissions

First off, check out this quick 30s video of the AI agent in slack demo: [link removed] Secondly, agent example of using an mcp connection through the app's agent. In this case it uses every.ai (my startup 😀) to make a formal proposal offer to a speaker: [link removed] I really think this will change your use of the platform, and allow you to embed speaker and conference…

@BrandonMChu

Visit ↗

SessionBuddy

sessionbuddy-development.shiny-cloud-dd47.workers.dev

Sign-up

Cloudflare WorkersCloudflare

Agents: Claude Code · Codex
Models: Claude Fable · GPT 5.6 Sol · GPT 5.6 Terra/Luna
Spend / usage: $500 reported

Process & submission notes +

Build process

Collected all the requirements from the doc file and the YouTube demos. Also went through other demos of session board and collected screen shots. then tasked the agents to create a comprehensive plan. Asked one agent to critique and prove the plan does not meet the requirements and other agent to disprove.…

Additions & omissions

While the UX needs a bit of makeup, the backend, namely api-design, db design, observability, debuggability, docs, crash-friendly long running and short running services, virus scanner for uploads, environment isolation, secrets management, and providence of the build and its deployment that we maintain, in other words all the boring and hidden but important stuff is built on…

Visit ↗

Greenroom

greenroom.greenroom.workers.dev

Open demo

Cloudflare WorkersNext.jsCloudflare

Initial inspection

⚠ Expected an open demo but the submitted route landed on sign-in.

Agent: Copilot
Models: Claude Fable · Claude Sonnet/Haiku
Spend / usage: $500 reported

Process & submission notes +

Build process

Built Greenroom, a SessionBoard-style CFP/conference organizing tool, iterating feature-by-feature with an AI coding agent: onboarding task catalog, admin task management dashboard, R2 file uploads, public widgets, admin content editing, and an AI-assisted review scoring feature (organiser/reviewer/speaker roles with an AI-generated suggested score + rationale reviewers can accept or override).…

Additions & omissions

Notable: the AI-assisted review scoring feature is intentionally "trust, but check" — it shows an AI-suggested score and rationale but always requires human reviewer confirmation, and it is explicitly text-only today (it does not evaluate speaker-shared video or attached files/CVs, which is flagged as a known limitation rather than silently ignored).…

Visit ↗

Session Canvas

sessioncanvas.dev

Open demo

VercelNext.jsSupabaseCloudflare Workers

Agent: Claude Code
Models: Claude Fable · Claude Opus
Spend / usage: $1 reported

Process & submission notes +

Build process

Financial spend field not working, answer; See attached. About half a Claude Max sub, so $100-$120. Tokens: 963.0M You will see on the attached PDF the model usage split and then the efficiency split with sub-agents. I find that intelligence/model matters for the upfront planning phase, where I will spend most of the time steering and understanding the brief/spec.…

Additions & omissions

Airtable integration outlined in the docs. You can export and sync Submissions. Sessions and Speakers for reference. Vercel hosting: [link removed] - Cloudflare hosting tested here: [link removed] - Tried to add to [link removed] but Alpha full!

@rjcrabbe

Visit ↗

Session Hero

session-hero.narralabs.com

Manual access

Tailwind CSS

Agent: Claude Code
Model: Claude Fable
Spend / usage: $50 reported

Process & submission notes +

Build process

I run a software agency do started with those lens. Understand the product, breakdown features, and let AI do the coding.

Additions & omissions

Third party CMS sync. Airtable or Hubspot or Salesforce.

@westoque

Visit ↗

Saas Killa

saas-killa.0xosprey.com

Demo login

Cloudflare

Initial inspection

⚠ Homepage returned HTTP 500 during capture.

⚠ Homepage displayed an application or runtime error during capture.

Agents: Claude Code · Codex · OpenCode · Pi / Oh-my-pi · Devin
Models: Claude Fable · Claude Opus · GPT 5.6 Sol
Spend / usage: $500 reported

Process & submission notes +

Build process

Used my meta harness + smithers.sh to scope and implement and test via browser use

Additions & omissions

See here: [link removed]

@0x_Osprey

Visit ↗

Session Party

sessionparty.com

Open demo

CloudflareCloudflare D1

Agents: Claude Code · Codex · Pi / Oh-my-pi
Models: Claude Fable · Claude Opus · GPT 5.6 Sol
Spend / usage: All subs, went through ~5 max($200) max plans that I had.

Process & submission notes +

Build process

Started from infra side with Cloudflare D1,R2, and DO with PartyKit(PartyServer or whatever its called now). Kept a command registry and layered up from there.

Visit ↗

ConfPilot

confpilot.com

Manual access

Cloudflare

Agents: Claude Code · Codex
Models: Claude Fable · Claude Opus · GPT 5.6 Sol
Spend / usage: AgentsView shows over 2000 based on token use but was using it through chatgpt max and claude each $200 plan

Process & submission notes +

Build process

It was fun. I was experimenting with long running goal in codex. Let codex be orchestrator and run claude design via codex (i think designs from ai looks good, but it converging towards similar design) Here is claude design used by codex using app use [link removed] And here is goal used on codex Bring ConfPilot to a verified, competition-ready v1 release from its current repository and worktree state.…

Visit ↗

smolboard: open-source speaker & CFP management

smolboard.app

Demo login

CloudflareReact

Agents: Claude Code · Codex
Models: Claude Fable · Claude Opus · GPT 5.6 Sol
Spend / usage: $3,709.20 of equivalent API spend. 6 days, 8,698 API calls, 3.95 billion tokens. I measured this from the Claude Code logs. I did not estimate it. Claude Opus 5 4,923 calls $1,528.09 Claude Fable 5 3,767 calls $2,181.11 Most of the cost is not writing code. Re-reading context is 72.7% of the bill. Writing output is only 6.5%.…

Process & submission notes +

Build process

I built smolboard with Claude Code on Pylon. One process runs the API, login, live sync, web pages, and the agent. It took 6 days and 199 commits. The app has 36 data types, 123 server functions, 34 pages, and 236 tests. I worked in three ways. First, I checked every screen in a real browser. I did not trust the code.…

Additions & omissions

Please look at these: - Content desk. Speakers upload slides and photos. Organizers review each version. Nothing reaches the public page until someone approves it. - Calendar invites that work. The .ics file is a real invite. You can accept it in Gmail and Outlook. If you move a talk, the speaker's calendar changes. It does not add a second entry.…

@ericcampbell59

Visit ↗

KMS

kms.r-s.workers.dev

Open demo

Cloudflare WorkersCloudflare

Agent: Claude Code
Models: Claude Fable · Claude Opus · Claude Sonnet/Haiku
Spend / usage: About $400 on APIs for the evals, one week of limits for Claude Max

Process & submission notes +

Build process

First stage Claude collated all the spec information from the Brief doc into a project plan. Then Fable managed the initial build by launching fleets of subagents for each milestone. I did an initial check of a test page but then hardly looked at it until it was finished. Then I got Fable to run the eval judge and fix issues in a loop, and in the meantime I was manually inspecting the app for usability and functionality.…

Additions & omissions

I tried to build all the features requested including all bonus etc. I didn't use Airtable as primary data source as this would have been too slow but added a sync. A possible feature I decided against was AI review assistance, it wouldn't have been high enough value. I added a lot of extra features, or unasked for stuff. Responsive: you can view most things on a mobile.…

@jim_ej

Visit ↗

Omotenashi

omotenashi-dev.dfiander.workers.dev

Demo login

Cloudflare WorkersReactCloudflare

Agents: Codex · Cursor
Models: GPT 5.6 Sol · GPT 5.6 Terra/Luna · Grok
Spend / usage: about 140 i didn't go crazy. upgraded to codex 100, 20 on grok, 16 api grok, 20 cursor. like i dunno half sub would be ok?

Process & submission notes +

Build process

first i thought about my experiences with similar software and crm tools in the past, i knew i didn't want a boring grid based tool. then i tool the google drive doc, and transcript from the youtube video, durected research and used sol ultra to build Beads (around a dozen epics and 105 beads). Visual mockups were done with grok (althought i tested kimi, codex, minimax, deepseek and grok i can send these to you for fun to look at).…

Additions & omissions

all bonus points were persued, crm maybe is a little light, maybe front end widgets too

Visit ↗

Event Manager OS

program-desk.ian-208.workers.dev

Open demo

Cloudflare WorkersCloudflare

Agents: Claude Code · Codex
Models: Claude Fable · Claude Opus · GPT 5.6 Sol
Spend / usage: Approximately 12.8M fresh-input and output tokens, including about 1.07M generated output tokens, $60 in Anthropic Credits and $40 in OpenAI credits

Process & submission notes +

Build process

We started with the eval kit, not the SessionBoard screenshots. The rubric showed that most clones cover basic CRUD but fail on permissions, scoping, and cross-role handoffs. That gave us a clear thesis: prove that one submission can move from Submitted to Published, with persistent evidence at every stage.…

Additions & omissions

Speaker CRM: I left cross-event speaker management out because it was extra credit and would have diluted the core lifecycle. Airtable persistence: I kept the existing data layer. Replacing it on the final day would have put the core workflow at risk for a modest bonus.…

@iankar_

Visit ↗

Callboard

callboardhq.com

Open demo

Cloudflare WorkersCloudflare D1Cloudflare

Agents: Claude Code · Codex
Models: Claude Fable · Claude Opus · Claude Sonnet/Haiku
Spend / usage: Two subscriptions (Claude Max 20x, ChatGPT Pro/Codex 20x) plus $99.92 in receipted API credits (receipt attached). Measured usage where runs printed totals: Codex ~50M+ total tokens across ~30 runs (largest single runs 14.4M and 12.0M, overwhelmingly cached input); Claude agent telemetry measured in the low tens of millions…

Process & submission notes +

Build process

One orchestrating Claude Code session ran the project like a small engineering org: every feature in its own git-worktree lane with a driver agent that briefed Codex, adversarially reviewed the diff, and had to show red-then-green test evidence plus mutation checks (break the fix, watch the test fail) before merge. Cross-family sweeps (GPT reviewing Claude's merges and vice versa) ran between waves and found real cross-PR seam bugs.…

Additions & omissions

Additions beyond the brief: an MCP server (mcp.callboardhq.com) and per-event llms.txt so agents are first-class users; HMAC-signed webhooks with a delivery log (optional Svix driver); a docs site (docs.callboardhq.com); an event-setup checklist for non-technical organizers; kanban board views; conflict gates that refuse to publish a physically impossible schedule (with…

@conorbronsdon

Visit ↗

Conference Ops

conference-ops-pilot-app.techfren.workers.dev

Manual access

Cloudflare WorkersCloudflareCloudflare D1

Agent: Codex
Model: GPT 5.6 Sol
Spend / usage: $200 for one max codex sub

Process & submission notes +

Build process

Gave inital video walkthrough and initial google doc to codex with sol ultra in a /goal to work with adversarial critiquing subagents overnight until my wake time (9am) then similar on the evals once they came out

Additions & omissions

Added a launch experience with starter templates/import, brand kit and live preview, readiness assistant, guided tour, Airtable status, and client-facing documentation. Added review plans/rounds, track-scoped assignments, blind review, recusal, applicant revision loops, speaker task/forms/files/comments/resources, content approval/version history, schedule views/conflict…

Visit ↗

OpenRostrum

openrostrum.com

Sign-up

ReactCloudflare

Agent: Pi / Oh-my-pi
Models: GPT 5.6 Sol · Grok · Any China open model
Spend / usage: >2B tokens

Process & submission notes +

Build process

My goal wasn't to manually drive the agents. I wanted to build a system that could automatically and autonomously build the project. So, I spent most of my time actually building the guardrails, the goals, and the software factory itself, rather than driving any agents. First task was to set a lot of static linting and custom rules so that the agents couldn't write slop code.…

Additions & omissions

My goal is to create an actual product that can be used to solve real people's problems (not "just a copycat"). I want to democratize access to software, we shouldn't charge crazy bucks for saas that can be vibe-coded. Additionally, the product is fully open source, so you can self-host it on your side too: [link removed]

@VDMatos

Visit ↗

ProgramFlow

programflow-evaluation.programflow.workers.dev

Demo login

Cloudflare WorkersPostgresCloudflare

Agent: Codex
Model: GPT 5.6 Sol
Spend / usage: $400 actual financial spend: two ChatGPT Pro 20x subscriptions at $200 each; no API billing. Account 1 used its original Pro allowance plus three banked rate-limit resets (four allowance windows total). Account 2 used its original allowance without a reset (one additional window).…

Process & submission notes +

Build process

I started by opening every source related to the challenge in separate tabs: the brief and docs, evaluator repository, walkthrough video, SessionBoard itself, and Discord. I asked Codex to use Computer Use to gather the complete context and separate useful questions and confirmed answers from noise. Copy-pasting everything would have been slow and messy, especially because many sources included images and screenshots.…

Additions & omissions

Beyond the core brief, I added Speaker CRM, strict role isolation, deterministic evaluator tooling, an Evidence Center, schedule-conflict repair, five public program surfaces, and inspectable integration history. I intentionally skipped Forge hosting and did not fake external services—live integrations require real credentials.…

@Vasu_ai123

Visit ↗

GrandStage

grandstage.everyai-com.workers.dev

Manual access

Cloudflare WorkersCloudflare

Agents: Claude Code · Codex
Models: Claude Fable · GPT 5.6 Sol · GPT 5.6 Terra/Luna
Spend / usage: I think the token spend is around $5,000 because I was measuring with a library that I made. It tracks all of my Claude Code and my Codex sessions. I have different ones, and I am posting a screenshot of it as well. This happened over the entire last week. It takes one week, and it's around $5,000 of API tokens.…

Process & submission notes +

Build process

Here is my process: 1. Started with studying the session board and seeing all the features. 2. Went through the documentation, all the documentation screenshots, and similar profiles and systems that are there. 3. Had a session with Fable 5, uploaded document screenshots and everything, and made a UX flow for the user. 4.…

Additions & omissions

Haven't had things that are super seriously skipped, but we haven't tested the A-table integration yet. We don't have an A-table Pro subscription, but we haven't tested the A-table integration properly. In our cloud testing, it said it will be working fine according to A-table docs. That is the only thing I want a bit of clarity on, and end to end we have tested all the things.…

@in @phanindra_ai

Visit ↗

Programcue

app.programcue.com

Demo login

Cloudflare D1ReactCloudflareCloudflare Workers

Agents: Claude Code · Codex
Models: Claude Opus · GPT 5.6 Sol · GPT 5.6 Terra/Luna
Spend / usage: $9,284.68 API equivalent cost, more detail in the proof upload. I used subscription so my real cost is probably around the $500 limit.

Process & submission notes +

Build process

ChatGPT Work to read the requirements doc and watch the videos, ChatGPT Chat with Sol Pro to come up with architecture, plan and initial designs. I decided to tackle a number of optional items as part of requirements. I built to those requirements initially with Sol Ultra in codex, fanning out subagents. Then follow on work I worked with multiple agents across different worktrees and running review/fix loops before fixes.…

Additions & omissions

- Cloudflare D1 is the default transactional database. Airtable may instead be selected as the authoritative event-data repository; authentication, audit, credentials, operations and other control-plane data remain in D1.…

@deavski

Visit ↗

Project Greenroom

project-greenroom-api.adityak6798.workers.dev

Open demo

Cloudflare WorkersCloudflare

Agents: Claude Code · Codex · Copilot
Models: Claude Opus · GPT 5.6 Sol
Spend / usage: $420 reported

Process & submission notes +

Build process

Initial plan -> Build harness to ensure good quality autonomous code -> run parallel agents in waves -> run against eval to check completeness -> run parallel agents -> ... (loop till satisfied)

Additions & omissions

Email provider, no Accelevents test (since I don't have an account).

Visit ↗

StageStack

stagestack.dev

Demo login

VercelReact

Agents: Claude Code · Codex · OpenCode · Cursor · Antigravity
Models: Claude Fable · Claude Opus · GPT 5.6 Sol
Spend / usage: I have the 200USD for Claude and Codex, I have the 15USD for Command Code, 10 for OpenCode Go and finally 60 for Cursor. The primary use and where I had to be smart with usage was Codex and Claude. 2 [person removed] resets really helped.…

Process & submission notes +

Build process

A lot of it can be seen on the git history of the project, but it all started here: [link removed] I created the business context file from general "search" with Gemini, got the challenge file to compile the spec and what needed to be done, I had gemini watch the youtube video and get the transcripts and as I added all of this to the repo, I had Fable keep it organized From there, I asked 5.6-SOL to interview me, question by question so…

Additions & omissions

I didn't build a lot of agentic experiences, kept it "admin ui" style, but with the extension added MCP which allows the use of your own agents to get things done. I focused on making this a real product, I'm quite confident you have what you need to run a conference on this as is.

@EquipeAI@MemphisBR

Visit ↗

Marquee

marquee.stage11.dev

Open demo

Cloudflare

Agents: Claude Code · Codex · Pi / Oh-my-pi
Models: Claude Fable · Claude Opus · GPT 5.6 Terra/Luna
Spend / usage: $500 reported

Process & submission notes +

Build process

I used C11, which is the Stage11 terminal multiplexer, as well as lattice, which is Stage11's agentically optimized replacement for linear.

Additions & omissions

Marquee is badass. We have every requirement, as well as these upgrades that I can think of off the top of my head: - **Agent-native.** For advanced users comfortable with Claude Code, Marquee is the ideal platform: a professional-grade skill file (served live at `/SKILL.md`), a 49-command CLI, and scoped credentials let your agent drive the entire surface first-class —…

Visit ↗

Sympose MVP

ethical-opera-murphy-somewhere.trycloudflare.com

Open demo

Next.jsCloudflare

Initial inspection

⚠ Temporary trycloudflare URL may expire without notice.

Agents: Claude Code · Codex
Models: Claude Opus · GPT 5.6 Sol · GPT 5.6 Terra/Luna
Spend / usage: ~$850 (I know y'all said you were only reimbursing $500 so no worries on the remaining balance!)

Process & submission notes +

Build process

spent Thurs brainstorming about what this could be and setting up infra (building up a more formalized agent coordination plane), weekend spent a lot of time thinking about primitives and how I wanted to approach this from ground up, started implementing Mon - Wed

Additions & omissions

I think I probably overfocused on architecture, but this means this product is super modular and easy to build off of. Should be fairly easy to extend whatever desired new features off of this. I don't have experience with Airtable / CRM systems so the connectors could be janky. Spent a lot of time on provenance and auditability.…

@statarbRS

Visit ↗

Trackstage

trackstage.app

Sign-up

ReactCloudflare

Agents: Claude Code · Codex · OpenCode
Models: Claude Fable · Claude Opus · GPT 5.6 Sol
Spend / usage: 1x Codex 200$ sub, 2x Anthropic 200$ – spent 4k$ in token equiv costs

Process & submission notes +

Build process

As mentioned above – 1 core Fable orchestration thread for everything that will manage subagents via claude`s native task tool, workflow tool & CLI calling of opencode & codex for in depth adversarial reviews. There is a slightly opinionated memory system to ensure that we will survive compactions & can sustain loads of subagents with shared context – the docs/ folder served that purpose & the exact work contract can be seen in…

Additions & omissions

This is performant, scaleable, prod-ready & was built in more or less 24h. Ill let the work speak for itself, but I am confident this is a winner and can replace Sessionboard reliabely :)

Visit ↗

Callboard organiser sign-in

callboard.eddyudotong.workers.dev

Open demo

Cloudflare WorkersReactCloudflare

Agent: Claude Code
Models: Claude Opus · Claude Sonnet/Haiku
Spend / usage: $382.76

Process & submission notes +

Build process

I started by watching your walkthrough on YouTube and read the Google docs carefully. You emphasized on Sessionboard being slow so I made sure server render time is now on every screen. You described it as a very fancy form builder, so that’s where most of the work went. I also did my best to ensure that publishing now runs a quick safety check first This was deployed to Cloudflare Workers as you suggested.…

Additions & omissions

The form builder will alert you if you’re about to publish a rule that stops people from applying. This idea came straight from your video, where a “minimum of two speakers” rule accidentally blocked the person who created the form. When different settings affect each other, the system combines them into one clear sentence that shows exactly what an applicant will see.…

@UEddy

Visit ↗

Podium

podiumstack.com

Demo login

Cloudflare WorkersCloudflare D1Cloudflare

Agent: Claude Code
Models: Claude Fable · Claude Opus
Spend / usage: $300

Process & submission notes +

Build process

I created a domain model first that captured the requirements. My goal was to create a factory to create SaaS from the domain model and it's easy to build new features by updating the domain model. I created specialized agents -- implementer (which was biased into the kind of stack we want to build), product-taste agent, red-teaming agent and performance management. I had a few skills to check for drift between code and domain model.

Additions & omissions

Event driven architecture - perfect for integration with other 3rd party apps Validated Performance & Security Factory approach - you can add new features by making edits to the domain model Agent can help with the admin operations (through plugin & APIs)

Visit ↗

OpenCallboard

opencallboard.com

Sign-up

Cloudflare D1CloudflareCloudflare Workers

Agent: Codex
Models: Claude Opus · GPT 5.6 Sol
Spend / usage: Approximately $250 USD: one $200 Codex subscription used throughout the build and approximately $50 of Bedrock Claude Opus usage for evaluation and review. I also used AI tools to produce the walkthrough video and purchased the project domain, but I am not including those in the reimbursement estimate.

Process & submission notes +

Build process

I began by consolidating the competition brief, organizer walkthrough, Sessionboard reference flows, and later Discord clarifications into one acceptance-oriented product model. I prioritized the complete vertical workflow over broad feature count: event/form setup -> public CFP -> track-based review -> acceptance -> automatic speaker/session/onboarding creation -> speaker portal -> real files/email/calendar -> agenda/conflict detection…

Additions & omissions

Notable additions: real email and calendar delivery; object-backed speaker files; self-serve organizer signup at app.opencallboard.com; role-scoped private access; Airtable sync; D1/R2/Queues persistence; public session/speaker widgets plus JSON, XML, and iCal feeds; API tokens and webhooks; mobile public views; conflict detection; open-source deployment and a real custom…

Visit ↗

vibeboard

vibeboard.io

Open demo

Cloudflare WorkersCloudflare D1ReactCloudflare

Agent: Claude Code
Models: Claude Fable · Claude Opus
Spend / usage: I have 200$ subscription spend around 2$ api equivalent, roughly 2B tokens

Process & submission notes +

Build process

Details in X article: [link removed] short version: I went all-in on pure agent mode: one person, one Claude Max subscription, three days, and a hard rule that I wouldn’t read a single line of code. Most of my time was spent before anything got built. I set up all the accounts and API keys myself so nothing would block the agent later.…

Additions & omissions

Built for agents too. MCP server at /api/mcp — six read tools with no key or signup, nine with one. Writes refuse unless you pass confirm:true, and hand back the plan they'd have run instead. Plus OpenAPI, a /agents page, /llms.txt. An API that speaks Sessionboard's dialect.…

@RRaphaell @RaphaelKalan

Visit ↗

Web Production Ae0c8

web-production-ae0c8.up.railway.app

Manual access

React

Agent: Cursor
Models: GPT 5.6 Sol · Grok
Spend / usage: $50

Process & submission notes +

Build process

Aim was to see if following requirements engineering principles will allow completion of project autonomously ([link removed] Got the agent to read the requirements engineering book, then, formally note them down, and note down all the flows. Then extract specifications from it. Each part of the requirment and specification was tagged and IDed. Then convert the requirements and specifications into Lean.…

@hargup13

Visit ↗

conference-engine

conference-engine.65labs.org

Open demo

VercelNext.jsPostgresCloudflare

Agents: Codex · Cursor
Models: Claude Fable · GPT 5.6 Sol · Grok
Spend / usage: I didn’t maintain a project-scoped token ledger, so I can’t provide a defensible token total without inventing one. The project was built mostly through existing Codex, Cursor, Claude, Grok, and Kimi access rather than metered API billing, so incremental cash spend beyond those subscriptions was approximately $0.

Process & submission notes +

Build process

I started by converting the buyer brief into a parity map and one domain spine: CFP → review → decision → speaker preparation → scheduling → publication. The product was implemented in bounded slices using parallel agent worktrees, with changes merged through unit, Worker, type, lint, build, and browser checks. I then dogfooded the real production lifecycle.…

Additions & omissions

Additions beyond the core brief include separate decide-versus-notify queues, a live program cockpit, multi-event ownership and teams, a lightweight cross-event speaker CRM, a per-event agent API with OpenAPI, Airtable and Accelevents outbound sync, embeds, CSV/XLSX/ZIP exports, calendar feeds, and schedule auto-placement.…

Visit ↗

DemoConf 2026

openevents.engineer

Open demo

ReactCloudflareCloudflare WorkersCloudflare D1

Agents: Claude Code · Codex
Models: Claude Fable · Claude Opus · GPT 5.6 Sol
Spend / usage: At least three concurrent 20× Max agent sessions or subscriptions. I do not have a reliable token or API cost total, so I am not asking for reimbursement. The app runs on Cloudflare Workers, D1 and R2 within free-tier limits.

Process & submission notes +

Build process

I read the full brief, screenshots and organizer walkthrough, then turned them into one working flow: configure the event, publish the CFP, submit, review, onboard speakers, send messages, build the agenda and publish the programme. I built it with React, Hono, Cloudflare Workers, D1 and R2. I worked one part at a time, with a real database path and browser test for each.…

Additions & omissions

Open Events covers CFP setup, submissions, reviewing, speaker onboarding, messages, files and scheduling. Reviewers get an assignment and scoring queue. Speakers get a private portal. The public site has the schedule, sessions, speakers, saved itineraries and iCalendar export. Orby is the support chat on the public site. It keeps one D1 conversation per visitor.…

Visit ↗

Podium

abstract-sync.lovable.app

Open demo

CloudflareVercelReact

Agents: Claude Code · Copilot
Models: Claude Sonnet/Haiku · Gemini
Spend / usage: Total Financial Spend: ~$170 USD (~15M–20M tokens total across models) Lovable Subscription/Credits: ~$100 — Platform tier usage for full-stack scaffolding, UI component design, state management setup, and continuous live deployment/previews.…

Process & submission notes +

Build process

• Architectural Strategy & Brainstorming: Passed the complete hackathon brief into Gemini to analyze requirements, map out core user flows, and structure the architecture. Refined and finalized the plan using Claude. • Frontend UI & Prototyping: Built and iterated the majority of the UI components, full-stack layout, and interactive dashboards using Lovable.…

Additions & omissions

• Post-Hackathon Iteration Offer: I recognize that hackathon MVPs might miss specific nuanced workflows tailored to your team's exact conference requirements. I am happy to dedicate up to 8 hours post-submission to build, refine, or adjust any missing features or integrations you need to make this production-ready for your upcoming event.…

Visit ↗

Tio Zeta

tio-zeta.vercel.app

Sign-up

VercelNext.js

Initial inspection

⚠ Homepage appears to be an uncustomized framework starter.

⚠ Submitter described this deployment as unfinished, unavailable, or fragile.

Agent: Claude Code
Model: Claude Sonnet/Haiku
Spend / usage: $0

Process & submission notes +

Build process

I broke my requirements down, since I have never really attended any tech events before it is a foreign concept to me. I used ChatGPT to break down the concepts and simplify the requirements. If there is enough time, I will append the chat links in the README file for your context and evaluation.

Additions & omissions

The program is aimed at technical users. I really tried my best to ensure the ease of use. By creating a custom DSL (Domain Specific Language) for form generation and conditional logic making it very scalable. When creating a form for speaker ensure to include the title field.

Visit ↗

Wejustlearn

wejustlearn.com

Open demo

Next.jsPostgres

Agents: Claude Code · Codex · Antigravity
Models: Claude Fable · Claude Opus · GPT 5.6 Sol
Spend / usage: Roughly 6.5 billion tokens total, about 90% of it on Opus and 10% on Fable. I did not pay that. Everything ran on a Claude subscription, so the $4,875 is what the same work would have cost at list API rates ($5/$25 per million for Opus 5, $10/$50 for Fable 5, cache reads at a tenth of input, cache writes at double on the one…

Process & submission notes +

Build process

This ran in two phases. The original deadline was three days, and those three were brutal, around fourteen hours a day, aimed squarely at getting a working MVP out the door. Then the deadline moved. I deliberately changed direction rather than adding more features: the next four days went into making it production ready. A lot got rewritten, and the UI and UX took most of that time. Before any of it I set up guardrails.…

Additions & omissions

The whole pipeline is here, not a slice of it. Open a call for papers, collect proposals, score them against a rubric, decide, chase speakers for bios and slides, build the agenda, publish it. You can run a real conference on this today. Things I added that I think matter: Speakers never get an account. Their email is their identity, a magic link signs them in.…

@in

↻ Consolidated resubmission

Visit ↗

Greenroom

session-bored.techwilliams-warren.workers.dev

Manual access

Cloudflare WorkersCloudflare

Agents: Claude Code · Codex · OpenCode · Pi / Oh-my-pi · Devin · Factory · Antigravity
Models: Claude Fable · Claude Opus · Claude Sonnet/Haiku
Spend / usage: i don't have a good estimate, fully saturating claude code limit @ 200/mo, about 50% codex. majority of claude usage was fable in the last few hours to polish. 9999999999999999

Process & submission notes +

Build process

firstmate orchestrator, driving 4-8x codex usually, glm for some implementations. user stories generated based on rubric and web exploration, periodic backlog generations with eval suite and user stories tickets

Additions & omissions

i'm sort of midway through the speaker CRM and have a bunch of parked issues for the deadline, idk if this is production-ready for a real event necessarily but some of it should be useful

@yikesawjeez

Visit ↗

Ruckus

ruckus.to

Open demo

Cloudflare

Agents: Claude Code · Codex
Models: Claude Fable · Claude Opus · GPT 5.6 Sol
Spend / usage: $300

Process & submission notes +

Build process

First off, I want to mention that this is my second submission because I rebranded with the deadline extension. The links in the original submission will redirect now, but were down for ~20 hours so I wanted to submit again just incase you hit a dead link. I started by giving the Competition brief to Chunky and had it run overnight to get started.…

Additions & omissions

Additions: Full CLI and MCP server (/api/mcp) so your agent of choice can drive things like the "AI generated schedule", instead of relying only on what's built into the app. Real email via Resend with per-recipient provider status, plus embeddable public widgets (sessions, speakers, agenda, itinerary, gallery, ICS/JSON/XML feeds) with an organizer accent override.…

@in@macheadley

↻ Consolidated resubmission

Visit ↗

Eventloom

eventloom.namuh.co

Demo login

Next.jsReactCloudflareCloudflare D1Durable Objects

Agent: Pi / Oh-my-pi
Models: Claude Fable · Claude Opus · GPT 5.6 Sol
Spend / usage: 20B tokens

Process & submission notes +

Build process

I started by turning the brief into a product contract and an end-to-end operating flow instead of cloning isolated screens. The source-backed core path is: publish a configurable CFP, collect and route submissions, let participants and accepted speakers manage profiles, tasks, forms and private files, run multi-round human review, build a conflict-checked agenda, prepare communications and calendar delivery, publish privacy-safe…

Additions & omissions

The most notable additions are a built-in Speaker CRM and an agent-oriented Eventloom CLI. The organization-scoped CRM connects contacts directly to program operations, with imports, tags, custom fields, segments, pipeline stages, notes, program history, duplicate detection, explicit merges and outreach through the same communications boundary.…

Visit ↗

SpeakerOps

speakerops.org

Open demo

Cloudflare D1TypeScriptCloudflare

Agents: Claude Code · Codex
Models: Claude Fable · GPT 5.6 Sol · Grok
Spend / usage: $1,000 reported

Process & submission notes +

Build process

Full video breakdown [link removed]

Additions & omissions

cli & sdk added for easy agentic administration and integration

@blockbrain_labs

Visit ↗

Fireside

onfireside.com

Demo login

Cloudflare

Agents: Claude Code · Codex · Devin
Models: Claude Fable · Claude Opus · Claude Sonnet/Haiku
Spend / usage: I used 2 Claude Max subs and 1 Codex Max sub. I did max out one of my Claude Max subs on just this project (weekly usage). Codex usage was large but I had usage resets. My out of pocket costs for this project were $176.87 spent on the sbek evals. Please reimburse that cost. Have receipts.

Process & submission notes +

Build process

(1) Plan before code. An object model, a ubiquitous-language vocabulary, 11 persona "hats," and an eval-parity map — done first, so the build had a contract instead of a vibe. (2) A gated autonomous build loop to the submission: build a feature, gate it with a fresh-context critic that had to pass before the next started, report between gates. (3) Eval-driven, two tiers.…

Additions & omissions

Additions we want buyers to notice: (1) A role-aware concierge — instant, zero-model answer chips for what the program already knows, plus an LLM path that by construction only ever sees public data. (2) An MCP surface — an agent can hold a speaker's whole obligation set: submit the talk, upload the deck, answer the green-room ping, all through the same guarded writes a human…

Visit ↗

Not on the program

curtaincallcfp.com

Manual access

Postgres

Initial inspection

⚠ Homepage returned HTTP 404 during capture.

Agents: Claude Code · Codex
Models: Claude Fable · Claude Opus · GPT 5.6 Sol
Spend / usage: The latest fleet report records approximately 3.68 billion Codex tokens across the constellation: 3,678,103,103 tokens. The hill climb opened an entirely new frontier for token burn because every product improvement could be captured, judged, turned into the next set of fixes, and measured again. I

Process & submission notes +

Build process

This was an amazing adventure. I'm running this in production already, and if any of this stack would be helpful to you, would love to help you and your team! We treated "kill this SaaS" as a research problem, then built a machine to grind it. 1. Field research first.…

Additions & omissions

Exports that make us a source, not a silo — ai.engineer-compatible sessions.json, speakers.json, calendar.ics, and llms.txt; a sessionboard-shaped API; and webhooks. Calendar files use stable UIDs so invites amend instead of duplicating when a room is assigned late. Goal is to make everything exposed through API, making the API I wish Sessionize and Sched had.…

@in@realgenekim

Visit ↗

Chautauqua Demo Org

chautauqua.cc

Open demo

ReactCloudflareSQLite

Agent: Claude Code
Models: Claude Fable · Claude Opus · Claude Sonnet/Haiku
Spend / usage: I spent 18.07B tokens for an equivalent API spend of $11,612. 17.52B cache reads (97%), 436M cache writes, 106.6M output tokens, ~0.4M uncached input — across ~208,000 API calls. By model: Opus $4,903 (planner, merge-trains, fleets), Sonnet $2,994 (swarm workers, runner loops), Fable $3,715 (interactive, meta-orchestrator…

Process & submission notes +

Build process

I used a cheap agent swarm implementation I mostly developed reimplementing Cursor's Agent Swarms blog post from July (my implementation actually appears to yield similar results but doesn't require a custom VCS or grok models and is cheaper on a API cost basis for the same SQLite implementation as described in the experiment) and incorporated [person removed] Thunderdome idea for when the merge queues get too backlogged.…

Additions & omissions

I say this after pulling and running about 10 of my most worthy competitors code: I think my software is better designed, more visually and logically coherent, and generally less of a slopfest than the rest. I hope you feel the same way.…

@gswangg

Visit ↗

Cicero

cicero-three.vercel.app

Open demo

VercelNext.jsCloudflare Workers

Agents: Claude Code · Codex
Model: Claude Fable
Spend / usage: probably close to 75% of a max 20x sub but most of that coming final day, no need to reimburse

Process & submission notes +

Build process

Writeup posted/linked from the submission and here . Will post in the discord too [link removed]

Additions & omissions

I compared feature sets across the full field of public/shared submissions, including my own submission. That survey is linked here and should be reproducible with a skill I built for this (skill text at /llms.txt on the field survey and in the source), could save y'all some time to use this [link removed] I also cover additions/omissions a bit in my submission writeup [link…

@EllAchE

Visit ↗

Youbii

youbii.com

Demo login

Stack not detected

Agents: Claude Code · Codex
Models: Claude Opus · GPT 5.6 Sol · Any China open model
Spend / usage: $200

Process & submission notes +

Build process

Spec -> Planning -> Implement

Visit ↗

Stagecraft

stagecraft.ai-c24.workers.dev

Open demo

Cloudflare WorkersTailwind CSSCloudflare

Agents: Codex · OpenCode · Devin
Models: GPT 5.6 Sol · GPT 5.6 Terra/Luna
Spend / usage: I used my codex account and opencode account. I think it'd be great if you gave out some openrouter or huggingface coupon codes. Thanks

Process & submission notes +

Build process

I spent time making prototypes and reviewing the brief to get a feel for the idea. The extra time helped a lot to polish features and testing. I've built a number of QA tools and the discord channel helped a lot with having a lot of clarity on the feature set and use cases.

Additions & omissions

Great initiative. Do this more and have more evals too. The concept is great