Build with agents.Keep your judgment.
Vyke is Luke’s public lab for using AI agents on real software work: the briefs, the guardrails, the failures, and the final human call.
Work
Real software, agents, and decisions shown in public.
Method
Scope it. Test it. Review it.
Point
Useful systems, not AI theatre.
Working in public
Evidence beats a polished demo.
Start with a real task
Use a live problem with clear constraints, not a benchmark designed to flatter a tool.
Give agents boundaries
Scope, tests, and review decide what an agent can do and what it must leave alone.
Keep the result inspectable
A useful workflow leaves evidence: the brief, the diff, the test output, and the decision.
Real work
Actual tickets and tools, not fabricated demos.
Receipts
Tests, failures, and source material stay visible.
Judgment
Agents help with execution; people keep the final call.
The operating model
AI can accelerate the work. It cannot own the consequences.
The system matters: instructions, narrow permissions, independent gates, and a human who can stop the run.
About Vyke
For builders who want the useful part of AI.
This is a practical record of using agents to build software: what worked, what failed, and what still needs a person in charge.
Scope
Agents get a bounded task, never a blank cheque.
Proof
Tests and review decide whether a result survives.
Judgment
A person keeps the power to accept, reject, or stop it.
Read
Writing starts when there is something worth publishing.
Essays will collect here after publication. No placeholder posts, invented dates, or fake authority signals.
Publishing status
No essays published yet.
Subscribe to the channel for the work in progress, not an invented archive.