# OpenAI Codex Projects at AI Tinkerers

> Canonical HTML: https://aitinkerers.org/technologies/openai-codex
> Markdown URL: https://aitinkerers.org/technologies/openai-codex.md
> Technology record last updated: 2026-02-23T02:30:10Z
> Generated: 2026-09-22T17:53:23Z

OpenAI Codex is a cloud-based AI agent that autonomously writes, debugs, tests, and proposes pull requests for software development workflows.

Codex is a powerful, cloud-based software engineering agent, powered by the specialized codex-1 model (a derivative of the GPT architecture). It streamlines the development lifecycle by autonomously handling complex tasks: writing new features, debugging code, running tests, and generating pull requests directly to a GitHub repository. Operating in a secure, isolated sandboxed environment, Codex ensures reproducible changes and can process multiple tasks in parallel, a key differentiator from sequential human workflows. Developers interact with it through the ChatGPT interface or the Codex CLI, delegating work in plain English (e.g., 'Find and fix a bug where the jump search algorithm doesn't handle empty arrays') to accelerate shipping and improve team productivity.

- Official technology site: https://openai.com/codex/
- Public AI Tinkerers demos and talks: 12
- Result page: 1 of 1

## Recent Public Talks and Demos

### [durable agent orchestration - or how i learned to stop worrying and love the waterfall](https://dubai.aitinkerers.org/talks/rsvp_gUMEdXOl6G4)

This demo showcases how we can improve one of the main issues with agentic coding: non-determinism, and loss of planning context and tasks due to context rot in complex projects. By using beads (https://github.com/gastownhall/beads) to track complex task planning in combination with prescriptive communication protocols and handover, we can not only keep a durable persistent state across multiple agent sessions in the same project, we also now have a consistent agent-agent communication protocol across coding agents, orchestrators, watchdogs, and dedicated task runners. In this demo we will demonstrate how beads works in conjuction with coding agents, an overview of its capabilities, and how it can drive autonomous agentic development work for multi-day goal runs. This is a showcase of how to use beads across various automation levels, including agent orchestration, spec driven development, as well as to create repeatable consistent plans that can be used by agentic tools

- Event context: AI Tinkerers Dubai - October Demo Day — 2026-10-10 — Dubai
- Public talk page: https://dubai.aitinkerers.org/talks/rsvp_gUMEdXOl6G4

### [Claude Code Forgets Between Sessions. I Built a Fix.](https://atlanta.aitinkerers.org/talks/rsvp_F2czpdWF3xs)

I built MLA, a context coordination layer that keeps Claude Code's working context consistent and up to date across sessions. The problem is simple: during one coding session, Claude discovers an important constraint, changes an architectural decision, or learns that an old assumption is wrong. Then you start a new session and that working context is gone. Teams try to preserve it in CLAUDE.md, documentation, memory files, rules, or other knowledge systems, but those systems eventually drift because staying current still depends on a human or agent remembering to make the update. I initially thought this was just my workflow. After talking with 30+ developers using coding agents, the same failure mode kept showing up. MLA maintains an active source of truth as coding work happens. It captures important decisions and discoveries, preserves where they came from, detects conflicting or outdated information, and keeps track of what is currently valid. When a new Claude Code or Codex session starts working, MLA gives it the relevant current context before it acts. I'll demo the full loop live: something changes during one coding session, the source of truth is updated, then a fresh session starts with an outdated assumption and MLA supplies the current information automatically. I'll also show the hooks, retrieval path, traces, and governance behind the handoff.

- Event context: AI Tinkerers Atlanta x AI Collective: Community Demos at ATL Tech Week — 2026-08-13 — Atlanta
- Public talk page: https://atlanta.aitinkerers.org/talks/rsvp_F2czpdWF3xs

### [What if loops are the problem?](https://nyc.aitinkerers.org/talks/rsvp_UEhibnXmrYg)

Loopbreaker is a public, local-first MCP server and visual SQLite review graph extracted from a deeper review-and-shipping system embedded in my application, Rordi. I built the original system after an AI code review reached thirteen passes and continued discovering new reasons not to ship. Loopbreaker isolates the reusable mechanism: an issue has a frozen set of enforced behaviors, evidence is attached to those behaviors, and review is limited to one comprehensive pass, one repair-verification pass, and—only when necessary—one shipping-decision pass. I’ll demo the working system live: clone the public repo, start its local MCP server, connect an AI coding agent, load a synthetic version of the thirteen-pass incident, and inspect the review graph. I’ll show the agent querying the acceptance surface, recording exact test evidence, completing a repair pass, and stopping automatically—while the visual interface separately explains whether the issue is actually ready to ship.

- Event context: NYC Summer Social: Rooftop Oyster Demo Day (two years of Tinkering in NYC) — 2026-07-29 — New York City
- Public talk page: https://nyc.aitinkerers.org/talks/rsvp_UEhibnXmrYg

### [pi-hydra: Mob Programming for Coding Agents](https://cologne.aitinkerers.org/talks/rsvp_WhIovGtreZ0)

pi-hydra is an open-source extension for pi.dev, a coding agent harness, that adds observer "heads". A head reviews the agent's work while the agent is still working. Each head is one markdown file with its own lens. The repo ships six as examples, but you write your own or tell the agent to write one. After every step a head makes an observation and decides: stay quiet, print a note for me, queue feedback for the next turn, steer the agent mid-run, or interrupt the run entirely. The author came to this from mob programming, one driver at the keyboard and several navigators watching the same screen. Most agent tooling points the other way and delegates work down to a worker that builds its own context and hands back a result. A head goes up instead. It sits above the driver, watches the same run, and can act on what it sees. It also does not have to rebuild context: each observation is a fork of the driver's own provider request, replayed byte-for-byte so it rides the prompt cache the driver already paid for. In the Anthropic sessions the author measured, one always-on head adds roughly 30% to total session cost, and a second head about the same again. A head sees everything the driver sees, so it inherits the driver's framing and blind spots too. It catches many problems while they are cheap, but it does not replace the independent review at the end. In the demo, an agent added a feature to an existing service while two heads watched. One caught a security hole the agent had inherited and never thought to check. The other caught the agent quietly swallowing errors in code it was writing at that moment. Both steered the fix in mid-run without derailing the original task. Three prompts, about two minutes.

- Event context: AI Tinkerers Cologne: World Cup Edition — 2026-07-15 — Cologne
- Public talk page: https://cologne.aitinkerers.org/talks/rsvp_WhIovGtreZ0

### [Hacking The Browser For Fun ~And Profit~](https://sf.aitinkerers.org/talks/rsvp_dZETAb_xF00)

We'll talk about XBOW's research into native application hacking: the agents' structure, objectives, and inner "iteration" patterns which worked to discover a few dozen high severity bugs in Chromium, including demoing an exploit of one vulnerability which has already been patched.

- Event context: AI Tinkerers San Francisco: Offensive Security Demo Night — 2026-06-26 — San Francisco
- Public talk page: https://sf.aitinkerers.org/talks/rsvp_dZETAb_xF00

### [Vibe Closing - Building a Virtual CRO](https://seattle.aitinkerers.org/talks/rsvp_sLyy5fqVjLw)

A virtual CRO. We used agents to do deep research across a company’s financials, CRM, and product data to identify which GTM experiments to run.

- Event context: AI Tinkerers Seattle: GTM Track — March — 2026-03-26 — Seattle
- Public talk page: https://seattle.aitinkerers.org/talks/rsvp_sLyy5fqVjLw

### [Replacing a consultant with an AI Council](https://columbus.aitinkerers.org/talks/rsvp_HPzao1xqrvI)

We had an outside consultant to write code on a legacy platform that none of our developers knew. AI was able to write some of the code, with mixed accuracy, and us being unable to verify it. But using a mixture of models, we were able to get something better than any individual model could produce. In this talk, I'll show how we played models from different providers against each other to sharpen and improve results.

- Event context: AI Tinkerers - Columbus March Meetup — 2026-03-02 — Columbus
- Public talk page: https://columbus.aitinkerers.org/talks/rsvp_HPzao1xqrvI

### [Reality Check: Building a Personal Fact-Checking Framework](https://tokyo.aitinkerers.org/talks/rsvp__DhnH1LfZlw)

Hot takes. Plausible theories. AI-generated slop. LLM psychosis. How do you keep track of what's actually true? Reality Check is an open-source framework for building a personal fact-checking knowledge base—using Agentic tools (Claude Code, Codex, etc) to track claims, sources, predictions, and argument chains with a LanceDB-backed semantic search. I'll show how the system extracts claims from articles, assigns evidence levels and credence scores, and generates analyses, but also how I went about building it, from an initial prototype to several refactors over the course of a few days.

- Event context: AI Tinkerers Tokyo - Toranomon Meetup - February 19, 2026 — 2026-02-19 — Tokyo
- Public talk page: https://tokyo.aitinkerers.org/talks/rsvp__DhnH1LfZlw

### [Bring Figma and Jira Context into Every Pull Request](https://nyc.aitinkerers.org/talks/rsvp_QlSzTLbdpSU)

I will be presenting Spec Reviewer, a coding agent built to ensure coding tasks are complete according to product and design requirements. I will show how the agent uses MCP to retrieve context dynamically, build a test plan, execute it in a sandboxed environment and finally interact with developers, PMs and designers in GitHub, IDE, CLI or UI.

- Event context: November Demo Day ft. Google Cloud and CopilotKit — 2025-11-17 — New York City
- Public talk page: https://nyc.aitinkerers.org/talks/rsvp_QlSzTLbdpSU

### [The fastest & most secure way to integrate AI into your  apps](https://boston.aitinkerers.org/talks/rsvp__8thBPULCPM)

An introduction to Apple's new on-device LLM — the Foundation Models Framework

- Event context: AI Tinkerers Boston June Meetup @ MIT! — 2025-06-30 — Boston
- Public talk page: https://boston.aitinkerers.org/talks/rsvp__8thBPULCPM

### [Exploraciones en Programación con IA](https://medellin.aitinkerers.org/talks/rsvp_4RzMvWV5LNo)

En esta charla, exploraremos las fronteras de la programación asistida por inteligencia artificial. Analizaremos herramientas revolucionarias como Codex Cloud de OpenAI y Claude Code de Anthropic, mostrando cómo estas tecnologías están transformando el desarrollo de software. Compartiremos ejemplos prácticos, discutiremos limitaciones actuales y vislumbraremos el futuro donde humanos y sistemas de IA colaboran en la creación de código más eficiente, seguro e innovador.

- Event context: AI Tinkerers Medellín #13 - SoftServe - 29 de Mayo, 2025 — 2025-05-29 — Medellín
- Public talk page: https://medellin.aitinkerers.org/talks/rsvp_4RzMvWV5LNo

### [ChatOne](https://seattle.aitinkerers.org/talks/rsvp_2pUjkIFnGVg)

ChatOne is an experimental platform that democratizes access to large language models (LLMs) by making them accessible through SMS and phone calls. Designed for regions and demographics with limited internet access, ChatOne ensures AI reaches underserved populations. This project explores how AI can learn from diverse cultures, reducing biases that arise when AI systems primarily interact with internet-rich environments.

- Event context: AI Tinkerers Seattle - December 2024 Meetup — 2024-12-12 — Seattle
- Public talk page: https://seattle.aitinkerers.org/talks/rsvp_2pUjkIFnGVg

## Related Technologies

- [Claude Code](https://aitinkerers.org/technologies/claude-code) ([Markdown](https://aitinkerers.org/technologies/claude-code.md)) — 216 public demos
- [MCP](https://aitinkerers.org/technologies/mcp) ([Markdown](https://aitinkerers.org/technologies/mcp.md)) — 129 public demos
- [Claude](https://aitinkerers.org/technologies/claude) ([Markdown](https://aitinkerers.org/technologies/claude.md)) — 174 public demos
- [OpenAI API](https://aitinkerers.org/technologies/openai-api) ([Markdown](https://aitinkerers.org/technologies/openai-api.md)) — 520 public demos
- [TypeScript](https://aitinkerers.org/technologies/typescript) ([Markdown](https://aitinkerers.org/technologies/typescript.md)) — 205 public demos
- [Amazon EC2](https://aitinkerers.org/technologies/amazon-ec2) ([Markdown](https://aitinkerers.org/technologies/amazon-ec2.md)) — 7 public demos
- [Anthropic API](https://aitinkerers.org/technologies/anthropic-api) ([Markdown](https://aitinkerers.org/technologies/anthropic-api.md)) — 66 public demos
- [Anthropic Messages API](https://aitinkerers.org/technologies/anthropic-messages-api) ([Markdown](https://aitinkerers.org/technologies/anthropic-messages-api.md)) — 2 public demos
- [AWS AgentCore](https://aitinkerers.org/technologies/aws-agentcore) ([Markdown](https://aitinkerers.org/technologies/aws-agentcore.md)) — 2 public demos
- [Beads](https://aitinkerers.org/technologies/beads) ([Markdown](https://aitinkerers.org/technologies/beads.md)) — 7 public demos
- [ChatGPT](https://aitinkerers.org/technologies/chatgpt) ([Markdown](https://aitinkerers.org/technologies/chatgpt.md)) — 83 public demos
- [ChatOne](https://aitinkerers.org/technologies/chatone) ([Markdown](https://aitinkerers.org/technologies/chatone.md)) — 1 public demo
- [Coordination graph](https://aitinkerers.org/technologies/coordination-graph) ([Markdown](https://aitinkerers.org/technologies/coordination-graph.md)) — 1 public demo
- [Gemini](https://aitinkerers.org/technologies/gemini) ([Markdown](https://aitinkerers.org/technologies/gemini.md)) — 188 public demos
- [GitHub](https://aitinkerers.org/technologies/github) ([Markdown](https://aitinkerers.org/technologies/github.md)) — 74 public demos
- [GitHub Copilot](https://aitinkerers.org/technologies/github-copilot) ([Markdown](https://aitinkerers.org/technologies/github-copilot.md)) — 20 public demos
- [GPT-3](https://aitinkerers.org/technologies/gpt-3) ([Markdown](https://aitinkerers.org/technologies/gpt-3.md)) — 191 public demos
- [GPT-4](https://aitinkerers.org/technologies/gpt-4) ([Markdown](https://aitinkerers.org/technologies/gpt-4.md)) — 529 public demos