Book a call

Reference

The method

All topics

Can AI agents build reliable production software?

Sterlon builds production software with AI agents, and every change is reviewed against its spec, has to pass a safety net of checks, and ships only when every check is green.

An orchestrator plans the work. Agents research, specify, build and verify in parallel. A safety net and a gate decide what ships.

Shown on the site: Method, on the front page

Reviewed: 2026-11-17

How does agent-based development work, step by step?

At Sterlon every change passes five stages: research, spec, build, verify and gate.

Research: the change is designed from the code that already exists, before anything is built. Spec: scope, owned files, rules and acceptance criteria are fixed in writing. Build: one small piece is built, together with its tests. Verify: the work is reviewed against the spec, and real defects are found and fixed. Gate: only green goes through; then a pull request opens, the one point where a person steps in, and only if the spec asks for it.

Shown on the site: Method, on the front page

Reviewed: 2026-11-17

Does a person review the code agents write?

Risky changes are approved by a person. At Sterlon the spec of each change says whether a person approves it.

Every piece of work gets two fresh review passes by agents before a person sees it. Where a system has real users, its bar includes a person reviewing every merge.

Shown on the site: Method, on the front page

Reviewed: 2026-11-17

What happens when an agent makes a mistake?

At Sterlon an agent’s mistake is caught in the verify stage, where the work is reviewed against its spec, or at the gate, which lets only green through.

In the example change on the front page, verify finds that a timeout is not retried; the defect is fixed before the change goes through the gate.

Shown on the site: Method, on the front page

Reviewed: 2026-11-17

Is the 5 to 30 times figure real, and what is it based on?

Teams working with this method deliver 5 to 30 times faster.

We measured 30 times when building new production-grade software, across several projects. In smaller teams we have measured 10 times. 5 times is our low estimate for larger teams building on an old code base. The result depends on the team and its present speed, so you get an estimate for your own team with the readiness assessment, before you commit.

Shown on the site: Adopt, on the front page

Reviewed: 2026-11-14

Is AI-written code disclosed?

AI agents write most of the code, and we say so before work starts.

Every change goes through the same gates and is recorded in a decision log delivered with the work. Risky changes are approved by a person.

Reviewed: 2026-11-17

What is level 3, orchestrated agentic development?

Level 3 is orchestrated agentic development: the developer and an orchestrator set what, why and how, and agents implement behind guardrails and gates.

At level 1, AI completes code and the developer still writes. At level 2, AI does whole tasks, fast, but the quality depends on who prompts.

Shown on the site: Adopt, on the front page

Reviewed: 2026-11-17

What does the developer still decide?

At level 3 the developer sets the what, the why and the how.

Decisions that are yours stay yours: spend, credentials, deploys.

Shown on the site: Adopt, on the front page, The areas, on the bar page

Reviewed: 2026-11-17

Which areas does Sterlon assess?

Sterlon assesses 10 areas: Code, Tests, Gates, Infrastructure and delivery, Security, Data protection, Operations, How the team works, Evidence and Agents in the product.

The bar page lists what each area holds, and which items apply to which kind of system.

Shown on the site: The areas, on the bar page

Reviewed: 2026-11-17

How is our quality bar decided?

Sterlon sets the bar from three things: what your reality shows is needed, what is reasonable to expect of a system like yours, and what the law and your contracts ask.

If nobody can give the reason for a rule, it is not in your bar.

Shown on the site: The bar page

Reviewed: 2026-11-17

Contact

Start with 20 minutes

Tell us what you are building, where your team is stuck, or what your production keeps telling you. You leave the call with a clear next step.

hello@sterlon.eu
Book a 20-minute call