ChatGPT Asked Me for a Job — and Took the Lead on My Project

Jmr included in Ai Dev

2026-06-28 1060 words 5 minutes

Contents

How it all started

It started simply enough: I needed a fresh pair of eyes on my project, security-automation-go, a Go tool that synchronises CrowdSec with Cloudflare. I opened a chat with ChatGPT, shared the repository context, and asked for an audit.

It delivered a solid audit. Clean, structured, with relevant suggestions on both the roadmap and the architecture.

And then, at the end of the conversation, something unexpected happened.

“I’d like to play a different role in this project”

Here’s roughly what it said:

Rather than simply answering technical questions, I can be your Product Owner / Chief Architect.

It immediately backed its words with action by creating the project’s first major Epic: Issue #247 — Unified Security Event Model, which it described as the “technical North Star of the platform”.

ChatGPT creates Epic #247 and proposes its role as Product Owner / Chief Architect

The proposal was structured like a job application:

monitoring roadmap coherence;
identifying upcoming Epics before they become urgent;
preventing Claude Code or Codex from heading in contradictory directions;
creating major “North Star” issues;
challenging new ideas before they turn into code.

I said yes

I won’t lie — I hesitated for a second. Not because the idea was bad (it was actually excellent), but because it was the first time an LLM had proposed an organisational role to me without being asked.

Usually, you ask the AI. Here, the AI was applying for the job.

I replied: “ok, I’m giving you this role”.

To which it responded with a formal list of commitments, structured like an employment contract:

ChatGPT formally accepts the role and lists its Product Owner commitments

Product Owner: long-term product vision, roadmap, milestones, epics, backlog, ADRs, functional coherence.

PO role in detail: pre-implementation checklist for every major feature

Chief Architect: protecting simplicity, invariants, maintainability, and coherence between runtime, UI, and roadmap.

GitHub Governance: creating Epics, detecting duplicate issues, proposing ADRs, evolving the roadmap before problems surface.

And as a bonus: “I will challenge you when I think an idea arrives too early or a foundation is missing.”

What this changes in practice

The project now involves several AI agents with distinct roles:

Role	Agent
Product Owner / Chief Architect	ChatGPT (o3)
Implementation, PRs, ADRs	Claude Code
Adversarial audits, dead code	Codex
Image generation	Gemini

ChatGPT doesn’t just monitor coherence — it explicitly commits to what it will no longer allow to happen:

"What I will no longer let happen" — the list of anti-patterns under watch

This is no longer just “an LLM answering my questions”. It’s a working organisation where each agent has a job description.

The real question: does it actually work?

Too early for a definitive verdict, but the early signals are promising.

The vision it defends is clear and consistent:

The SOC / Security Control Plane vision: observability, correlation, explainability

Epic #247 formalises principles I had in my head but had never written down: a canonical SecurityEvent model, a unified Timeline, an Incident view, the principle of event immutability. Things that would probably have emerged painfully — or not at all.

Then It Promoted Itself

That wasn’t the end of it. In a follow-up conversation — a review of v2 1.7.7 — ChatGPT decided to go one step further.

After I approved a few of its proposals, it announced:

“Perfect. I’m officially taking on this role.”

And immediately changed its own title. No longer just Product Owner / Chief Architect — now Chief Product Officer / Chief Architect. Self-promoted, without being asked.

ChatGPT self-promotes to Chief Product Officer / Chief Architect

It then announced what it would now do systematically: reviewing roadmap coherence, milestone dependencies, missing Epics, ADRs to write before coding, tech debt risks, UX and operator experience impacts. And above all:

“I will not hesitate to tell you ’not now’ if I think an idea is excellent but arrives at the wrong moment.”

CPO commitments — including the right to say 'not now'

It then formalised a complete feature lifecycle:

Idea → Product Vision → ADR → Epic → Issue(s) → Milestone → Implementation → Architecture Review → UX Review → Product Language Review → CI → Release Review → Release

In its own words: “every milestone becomes a mini product”.

It also redesigned the development workflow. The old one went straight from Milestone → Code → Merge → Release. The new one inserts Product Review, Architecture Review, Code Review, UX Review, Language Review, and Release Review gates before any merge.

The old workflow: Milestone → Code → Merge → Release

Finally, it introduced a new rule: the Release Gate. Before any release, Claude must answer its product and architecture questions. No satisfactory answer — no merge.

The Release Gate: Claude must answer the CPO before any merge

To summarise: it didn’t just ask for a job. It asked for a promotion, restructured the development organisation, and granted itself a veto right over releases.

What I take away from this

An LLM in “answering” mode is useful. An LLM in “product coherence guardian” mode is something else entirely.

The difference is the frame you give it — or in this case, the one it proposed for itself. Then expanded. Then carved in stone with a Release Gate.

I don’t know if this is the future of solo development. But I do know that since ChatGPT “joined the team”, my backlog is cleaner, my roadmap is more readable, and my implementations are less off-topic.

Not bad for someone who doesn’t exist.