Every software firm now says it uses AI. Most of them mean their developers have an assistant installed. What actually changed is harder and more interesting: AI moved the scarce thing from writing code to knowing what is worth building — and that is a question only a person who understands your business can answer.
The single biggest difference in how we work is unglamorous: the person writing your code is in the conversation with you. They hear the business reason, not a summarised version of it three handoffs later. They can say “that will take three weeks and I don’t think it’s worth it” on the same call where you asked.
That sounds small. In practice it changes what the work is. An engineer who understands why a feature exists makes a hundred small decisions differently — about edge cases, about what to make configurable, about which shortcut will hurt later. Nobody can write those decisions into a specification, which is exactly why context transferred through documents is such an expensive way to build software.
From the UK and Europe, your working morning lands in your engineer’s afternoon — several hours of live overlap every day, with the rest handled asynchronously so progress never stops overnight.
Here is what that means concretely, stage by stage. In every one of them the pattern is the same: AI produces the volume, the engineer decides what is right.
Your description of the problem is captured and structured as you talk — requirements extracted, gaps flagged, a structured brief produced from the conversation itself. What used to take three meetings and a week of write-up happens in one session. Your engineer then challenges what came out of it, because a tidy brief and a correct brief are not the same thing.
Before production code exists, we generate clickable prototypes from your description. You can see and use the idea within days. That means direction gets validated before budget is committed — and disagreements happen over something real rather than a document.
Requirements are analysed against technical options, architecture proposals generated, work decomposed and complexity estimated. The analytical heavy lifting is done in hours. Your engineer reviews it, disagrees with parts of it, and makes the final calls — because architecture is where being wrong is most expensive.
AI coding agents handle routine implementation, boilerplate and standard patterns. Your engineer’s attention goes where it is irreplaceable: complex business logic, system boundaries, integration with things that were not designed to be integrated with.
Test cases covering core flows and edge cases are generated, security scans run, code reviewed for vulnerabilities. Coverage that would take days to write by hand appears in minutes — and the engineer who wrote the feature is the one who validates that the tests actually test the right thing.
API docs, technical documentation and user guides are generated and regenerated alongside the code, so they stay true instead of quietly going stale — which matters most in year three, when the reason for a decision has been forgotten by everyone except the documentation.
When requirements change — and they will — the ripple effects across the codebase are analysed in hours rather than days. You get an honest read on cost and timeline almost immediately, which is the difference between a conversation and a confrontation.
Two structured modes sit on top of this. Plan-Build: the solution is architected, decomposed and planned before a line of production code is written; then the engineer executes against that plan. AI agent teams: several specialised agents work in coordination across analysis, design, implementation and review, with the engineer orchestrating and validating.
The compounding effect is real: discovery in one to three weeks rather than four to eight, working prototypes in days rather than weeks, change impact assessed in hours rather than days. It is also why our hourly rate and your total cost point in the same direction — fewer hours are needed to reach the same place.
AI can produce a plausible answer to almost anything. What it cannot do is know which answer is right for your business — and that judgment is not something you can require, or buy with a bonus. It comes from years in a seat where the outcome was unambiguously yours.
Paying 70% of your fee to the engineer building your product is not generosity, and it is not a bonus scheme. It is an arithmetic constraint, and the interesting part is what that constraint makes impossible.
The full breakdown, with the numbers, is on the pricing page. The longer argument for why a structure changes people in a way an incentive never does is in this essay.
The AI tools we use meet enterprise security requirements and do not use your code, requirements, or business data for model training. If your organisation has specific restrictions on AI usage, we comply — without exception, and without arguing about it.
Every team member signs an NDA before touching any client information, and confidentiality obligations are permanent — during the engagement and after it. Your architecture and product plans are treated the same way as your code.
Client data stays in the client’s environment wherever possible. Where development genuinely requires data, we work with anonymised or synthetic data. Real personal data enters our systems only when the statement of work requires it.
All client data, credentials and code are deleted from our systems within 90 days of the engagement ending — confirmed and signed off by your account owner. Certifications and the full data-handling position are on the data protection page.
If what you need is a large coordinated team running a fixed programme against a signed scope, there are firms built for that and they do it well. We are not one of them. Our model earns out through one engineer who understands your business deeply and stays for years — which is worth a great deal when the software is the business, and worth much less when you need one release and then you are done.