Skip to main content

Agent Setup and Oversight

The agent works from the repository: the conventions, the expert subagents for each domain, the tools it may call and a plan that shows a wrong assumption on paper. Every report gets checked against the build, the tests and the running system. What runs unattended and what waits for a person is set per environment.

Test, Approve, Release

Every change is built into a second slot, blue, and tested there with end-to-end tests. Green, the version your users see, takes that tested version once a person has approved it, and a rollback takes one step. You get every application in source code, backups from day one, and a log of every conversation.

Operations at a Glance

Your operation sits in one dashboard: which application runs, whether it answers, how it is configured and what the work log records. Behind it sit the CI/CD, from the GitHub Actions pipeline to the blue/green deploy, and one store for configuration values and keys that every deployment reads from.
Full conversation log
every conversation with the agent is kept alongside the code it produced
Your sign-off decides
a draft waits for your review before anything saves to a live system
Invoice traces back
every figure on the invoice traces back to a real, recorded work session
Site confirms live
the public site itself confirms an update before it counts as live

From Plan to Reliable Operations

Stocktake & Plan 3 steps
01

The Business Written Down

Before the AI works on its own, the knowledge of the business is written down where every session reads it: what runs today, where, which databases exist and how logins work. An open question stays marked as open, and a password is described by where it is kept.
02

Work Split Into Roles

The work gets split up like a small team: one role writes, one checks, one translates, one researches, one publishes, each with its own brief. A check decides whether something needs a second pass, at most two, so the AI's full attention goes where judgement is needed.
03

Plan First, Then Act

Spec-driven development gives every task the same written shape before it starts, so a wrong assumption shows up on paper before the build. The same discipline applies to a support request: the code gets looked at first, so the request names the real file.
Knowledge Built In 3 steps
04

The Harness at Work

The build runs under the harness from Harness Engineering: specialists for each area, an automatic check of every change from three angles and house rules the AI has to follow. Every instruction lives in exactly one place, and the work runs reliably because of it.
05

Recurring Work as a Button

A recurring task can go to the business side as a simple button in Microsoft 365. A mistake made once becomes a rule the AI keeps from then on, such as a backup copy made automatically before domain records change.
06

Rules That Enforce Themselves

Hooks enforce the house rules automatically, with a check at the end that nothing slips through. A long-running version can measure the website's load speed on its own, on a schedule, with tight limits on what it may touch, so nothing risky happens unattended.
Proven in Daily Operation 2 steps
07

Two Tools in Daily Use

Two tools built this way run here every day. One tracks working hours and turns them into billing. The other holds every password and setting in one safe place that every other system reads from automatically.
08

Every Update Proves Itself

Before an update counts as live, it has to prove itself: what runs gets compared against what was meant to go live, and the last check is whether the website answers. A deploy counts as finished once the host answers, a check that catches a site that stays unreachable while every container reports healthy.
Visible and Billed 2 steps
09

The Whole Operation at a Glance

A dashboard shows the whole operation at a glance: which app is running, whether it's reachable, how it's configured, and what the work log records. It runs in daily use itself, in the same look as the public website.
10

Test Honestly, Bill Honestly

The test report shows what got tested and marks every untested part as a gap. Billing is built from the recorded working hours, so every figure on the invoice traces back to a real session.

Frequently Asked Questions

What is agentic coding?
Coding agents such as Claude Code and GitHub Copilot plan, write and test the software themselves, working from a written plan under a harness of house rules. A person sets the goal and approves every update before it goes live.
What is spec-driven development?
Every task gets a written specification before an agent starts: what gets built, what it is checked against and which assumptions it rests on. A wrong assumption shows up on paper before the build starts.
Can a model write directly to a live business system?
The model proposes a draft as a form in the chat, prefilled with its proposal, and the save runs when the person looking at the form confirms it. Read-only connections stay read-only, and a skill points to the file that holds a value.
How do we know an update is live?
Each deploy compares what runs on the server with what was meant to go live and finishes with a check that the public website answers. The same check can run at any time.