- Matt Pocock argued that coding agents become trustworthy inside strict automated limits, such as types, linters and tests, that a human insists on.
- He said agents can write those rules and speed up the tests, but a human must care enough to tell them to.
Matt Pocock, creator of Total TypeScript, argued that AI coding agents become trustworthy when a human first insists on strict limits around them. He said so on NetworkChuck, in an episode published on 20 November 2026. The limits are lint rules, tests and type checks. He urged fixed automated checks over an agent's own judgement. What it takes, he said, is someone who cares enough to demand them.
What was said
Pocock described a linter as a program that checks code for bad patterns. One rule in his coding standards bans the any type in TypeScript. He called it a cheat, because it dodges the type system and weakens later checks.
He also described a tool called retro. When a code review turns up a problem, retro looks for an automatic way to catch it next time. Each run tightens the checks. Eventually, Pocock said, the agent is boxed in: "They literally cannot make a mistake."
In his account, agents do the building. He said they are good at finding lint rules to constrain themselves and at making tests run quickly. The human's part is to tell them to. He put it this way: "you can get the agents to care. It's just you need to care yourself."
He applied the same logic to choosing between a rule an agent reviews and a lint rule. When you have the choice, he said, pick the deterministic one, which runs the same way every time.
Why it matters
Our reading: agents write code faster than people can read it, so any weakness in a codebase can spread faster too. Without trust in agents, Pocock said, people end up checking everything and cannot scale beyond about four agents at once.
For managers, that moves the question. Safe agent output depends on the quality of the checks, not only on the model. Since agents can write the rules, the scarce input is someone who asks for them and owns the process. Pocock's account suggests that someone has to choose to care.
The other side
Pocock was open about the limits of his own setting. He said he builds personal software, not medical or finance applications, so he can trust his agents. In banking, he said, he would look hard at one-way doors and spend heavily on verification until he trusted them.
He offered no outside data on how well strict checks work. His line about agents not making mistakes describes a goal, not a measured result. Caring also remains open: if an agent does not know to do something, he said, it will not care about it either.
Written by the WebPulse Newsroom with AI assistance, and checked by our editorial review: every quotation was verified against the recording's transcript. How we use AI.
The conversation this talking point comes from
- NetworkChuck: How to Actually Code with AI (ft. Matt Pocock) (2026-11-20)





