Ben Chan Tech · live operating experiment
How far can one person push AI execution on a $20 monthly AI bill?
Ben Chan Tech is being rebuilt as an AI-native company with ChatGPT Plus as its only required AI operating expense. Ben sets direction, retains judgment and accountability, and delegates as much ordinary cognitive and execution work as the tools can responsibly carry.
The operating constraint
One required AI expense.
The core company must remain buildable, publishable and operable with one ChatGPT Plus subscription. Paid AI APIs, agent platforms, vector databases, AI CMS products and additional AI subscriptions may be tested separately, but none may become required infrastructure for the public knowledge surface.
The live experiment
The company is the case study.
The question is not whether AI can produce a demo. The question is how much real company work can move from human execution to AI execution while human judgment, provenance and responsibility remain intact.
- Captain minutes per useful shipped outcome
- Interventions caused by judgment, authority, missing context, failure or tooling
- Work completed in ChatGPT before specialist coding work is required
- Claims and pages traceable to real evidence rather than synthetic authority
Operating model
ChatGPT is the primary cognitive workforce.
Research, synthesis, simulation, planning, writing, content architecture, evidence mapping, review and much of site maintenance should happen inside the subscription first. GitHub holds durable state. The public site prefers deterministic artifacts. Specialist coding tools are used only when runtime engineering materially requires them.
If the final artifact can be produced in ChatGPT and stored as ordinary code or content, do not add another paid AI dependency.
Evidence, not synthetic authority
Built from accumulated human evidence.
The experiment starts with decades of software and music work, real professional cases, product experiments, essays, transcripts, technical systems and published methods. AI can search, pressure-test and recombine that record. It cannot manufacture Ben's experience, authority or conclusions.
Developer Forward
Developer Forward continues as an evidence surface.
The full Developer Forward offering is discontinued for now. Developer Forward Lite remains available as a free, deterministic five-case experience. The broader Developer Forward name remains the home for developer-judgment research and a possible future traditional course, but there is currently no paid destination or upgrade promise.
Questions the company is trying to answer
Questions this experiment should be able to answer with evidence
- Can ChatGPT maintain a production company website through GitHub?
- When is a $20 consumer AI subscription enough for real business operations?
- What company work still requires human judgment after execution becomes cheap?
- When does a coding agent add leverage, and when is chat enough?
- Can synthetic simulations improve an evidence corpus without becoming synthetic authority?
- How much human review does AI-generated public knowledge actually require?
Review routes
Two rooms are built for current reviewers.
Standing principle
Give AI as much room as it can responsibly carry. Keep the boundaries that matter.