The agents see the work.
Humans make the decisions.

A software agency being built from the ground up — organisationally, not just technically — for a world where engineering stopped being the bottleneck. We know what changed. We don't yet know everything it changes.

Day 3 / No clients / No case studies / Nothing to sell you yet

01 — The premise

Every software agency that has ever existed was organised around one scarce thing: engineering time.

The account manager protected it. The delivery manager sequenced it. The business analyst translated into it. The estimate priced it. The change request defended it. The org chart, the rate card and the statement of work are all scaffolding built around a single constraint.

The constraint moved.

We don't think that makes agencies obsolete. We think it makes most of their structure obsolete, and we would rather find out what replaces it by building one than by writing another post about it.

So this is the whole thing, from the ground up: what the roles are, what you buy, what it costs, who is accountable, and what it is actually like to be a person doing this work every day.

02 — What we actually don't know

These are not rhetorical. We don't have the answers, and we are suspicious of anyone currently selling them.

If writing the code is no longer the expensive part, what is the client actually buying?
What replaces the estimate? If you can see the working software before you commit, do you pay to go to production rather than paying for the sprint?
What is a product manager when the specification is executable?
Who is accountable for a system that nobody typed?
How do you run acceptance testing on something produced faster than a person can read it?
Where does the next generation of senior engineers come from, if the junior tier that grew them no longer exists?
What does taste mean when volume is free?
Do engineers still want the job? What is it now — author, reviewer, director, or something we don't yet have a word for?
What is the honest rate card, and would you publish yours?

03 — How we're going about it

We're going to spend a long time asking before we spend any time selling.

In public, on the record

Streams and a podcast with people who run agencies, buy from agencies and work inside them — including the ones who think this premise is nonsense. The discovery is the content. It isn't marketing for a thing that already exists.

Building, not describing

Our own tooling for acceptance, review and deployment gates, built in the open with the failures left in. If we can't make it work on ourselves first, we have no business charging anyone for it.

Clients last, not first

We are not taking on client work yet, and we'd rather say so plainly than run a waitlist we can't service. When we do start, we'll publish what we charge and how we arrived at it.

Radical honesty is an easy thing to put on a landing page, so here is ours: we have no proven model, no client list and no product. We have two engineers, a domain name and a strong suspicion.

04 — Who

James Hartt

Lead Software Engineer, agentic engineering · Hove, UK

James isn't theorising about agentic engineering. He's been doing it in production and counting: 874 commits across three production codebases in six months, around a quarter of his team's total output, and he wrote the AI agent guidelines the rest of the team now works to.

Fifteen years in the parts of the industry where shipping badly is expensive and immediate — nearly six years leading development on a cloud live-broadcast platform, iOS for HM Revenue & Customs, and a news app read by five million people.

The side projects are the argument rather than a hobby. A C++ WebTransport real-time audio server, written across two Sundays, so his guitar quartet could rehearse from different rooms. A 52,000-line Go platform with 146 end-to-end tests and 40 migrations that runs the fixtures for a Brighton parks tennis league. Both solo, both in real use by real people, both alongside a full-time job — which is roughly the point.

First-class BEng in Electronic & Electrical Engineering from Leeds, top of his year, holding the only award the department gives to undergraduates — voted by staff rather than calculated from marks.

tutti.rocks jim.tennis GitHub LinkedIn

Abrar Mahmood

Lead Software Architect · Co-founder & CTO, GradeDrive

Abrar has spent nine years working on the seams where software actually fails: ownership, architecture and the client conversation. Seven of them architecting a global retail commerce platform — point of sale, payments, transaction flows, inventory, and ERP and OMS integrations. The systems where being wrong costs money in public, instantly.

He led distributed teams and walked clients through high-stakes technical decisions, which is precisely the role this whole thesis puts under the most pressure. He rebuilt that platform around modular patterns that cut running costs and raised deployment velocity, and got environment onboarding down by eighty per cent.

He is currently co-founder and CTO of GradeDrive, an AI marking platform he took from zero into real classrooms in UK secondary schools. So he has already had to answer "what is this worth, and who is accountable when it's wrong" for money, not in theory.

He cares about clean system design, clear ownership, and creating leverage through good engineering — the three things an agency built this way will either get right or die of.

05 — Come and argue with us

Everything below goes to a real inbox that two people read. There is no sequence, no drip, no CRM.

Come on the show We want people who have run an agency, bought from one, or been chewed up by one. Especially if your experience contradicts ours. Pitch yourself → Follow the build Streams, episodes, and what we get wrong. Low volume, no marketing, and we'll tell you when it stops being interesting. Get the updates → Tell us we're wrong If you think the premise is nonsense, we would much rather hear it now than in two years. The strongest disagreement gets a whole episode. Take the swing → Work with us We're not taking clients yet and we won't pretend otherwise. This is the list for when we are, and you'll hear what we charge before you hear a pitch. Join the list →