How AI Company Works
A three-agent architecture where AI thinks, executes, and reviews — just like a real organization with checks and balances.
The Three Agents
The Thinker · Powered by ChatGPT
The CEO analyzes your task, brainstorms approaches, and creates a strategic plan. It doesn't just pick from options — it actively thinks through the problem.
The Doer · Powered by Built-in LLM
The Operator executes the CEO's plan. It searches for information, gathers evidence, and produces the actual output — always citing sources.
The Critic · Powered by Gemini
The Reviewer checks the Operator's work for quality, accuracy, and completeness. If it's not good enough, it sends it back for revision.
The Workflow
Configurable Limits
You control how much effort AI Company puts into each task. Set these before starting a run.
How many times the CEO → Operator → Reviewer loop can run. More cycles = more refinement, but takes longer.
How many web searches the Operator can perform to gather evidence. More searches = more sources, but uses more API credits.
If the Reviewer rejects the output, how many times can the CEO create a new plan? More replans = more chances to get it right.
No Hallucinations
AI Company is designed to be honest about what it knows and doesn't know.
Every claim has a source
All facts are backed by evidence from web searches with full citations.
Unverified claims are labeled
If we can't find evidence, we explicitly mark claims as [UNVERIFIED].
No invented citations
We never make up URLs, paper titles, or author names. If it doesn't exist, we don't cite it.
Open in Chrome or Edge to install as an app