How we work
Senior product engineers, AI doing the heavy lifting, and a release in weeks.
The software development lifecycle has not changed. Understand the problem, build the thing, test it, ship it, look after it. What has changed is that a small team of experienced engineers orchestrating AI can now do in weeks what used to take a much larger team months.
The shape of a project
- Before we start
A conversation first
You tell us what you are trying to do and why. We ask the awkward questions early: what happens if this is late, who uses it on a Monday morning, what does the current system get right. If we are not the right fit we will say so.
- Discovery
Understand the problem properly
We put our team and our AI tooling on the same questions. Who has this problem, what it costs them, what the current way of working gets right, and what has to be true for the new system to succeed. Our people run the conversations and sit with the users. AI works through your documents, data exports, support history and the market around you, and surfaces in hours what a person would take weeks to find. By the end we can state the problem in a paragraph you would put your name to, and there is a working skeleton you can click around.
- Specification
A live plan the AI builds from
The specification lives in Linear, the project system we have built our whole process around. Every requirement becomes a precisely written, testable piece of work, linked to the outcome it serves, the data it touches and the work that has to happen first. Our AI tooling is connected to it directly through the Model Context Protocol, a standard that lets AI systems read and act on the same records people do. The plan is the live set of instructions the AI builds from, and you can open it at any time and see exactly what is being built, why, and how far along it is.
- Build
Build in the open
You see working software every few days. The product engineers on your project make the technical decisions, set the guardrails and verify what comes back. AI does the implementation, the tests and the documentation, and updates the plan as it goes. Scope changes are a conversation, and they happen the same week.
- Release
A quiet day
By launch the system has been deployed dozens of times. Monitoring, error tracking and backups are already running. Going live is a configuration change.
- Afterwards
The same people, still on it
The team that built the system is the team that supports it. Ongoing work runs on a simple monthly arrangement, or we hand over cleanly to your team with documentation that describes what was built.
Product engineers and AI
The split that makes the speed safe.
We call the people who run our projects product engineers. The same people work out what should exist and build it, two disciplines the industry has kept apart for twenty years, so nothing is lost between the two.
What the product engineers do
- Decide what to build, and what to leave out
- Choose the architecture, data model and infrastructure
- Set the constraints the AI works within, and verify what comes back
- Own security, access and data handling
- Talk to you directly, every week
- Take the call when something breaks
What AI does
- Drafts the implementation from the written spec
- Writes the tests alongside every change
- Writes migrations, clients, fixtures and glue
- Keeps documentation in step with the code
- Reworks large changes in minutes when a decision moves
- Handles the volume, so the product engineers can handle the judgement
What does not change
The parts we will not compress.
The time saving comes from removing the waiting between stages. These six things still happen on every project, whatever the deadline.
Verification
Every change is tested automatically and reviewed by a person where the risk lives: the data model, the security boundary, anything touching money or personal information. Changes too big for that review to mean something get split.
Tests
Written with the feature, run on every commit. A red build blocks the merge, whatever the deadline.
Security
Secrets in a secrets manager, least-privilege access, and encryption at rest and in transit. Standard measures, all of them in place.
Observability
Logging, error tracking and uptime monitoring are set up before launch, so we hear about problems before you do.
Ownership
Repositories, cloud accounts, domains and data sit in accounts you control from the first day.
Plain technology
Widely used, well-documented platforms that any competent engineer could pick up.
Commercials
A number you can plan around.
Wherever we can, we scope a first release to a fixed price and a fixed date. Ongoing work after that runs on a straightforward monthly arrangement that you can stop with notice. We will give you a figure after the first conversation, with no paid discovery phase before it.
Because a project is weeks rather than months, the number is smaller than you are used to hearing from an agency, for the same software.