LTS Commerce Cobuild

Your developers are already using AI. The output is almost right, and that's the problem.

I spend a day a week for five weeks inside your codebase while your team builds something real with agents, and I make sure what comes out is code you'd be happy to own. You keep the tool. You keep the people who can build the next one.

Book a 30-minute call →

Five days of your team's time, eight days of mine, from £6,000 + VAT. Remote across the UK, or on site.

You've probably got past the easy bit

Autocomplete works. Everyone's using it. That was never going to be the hard part.

Where it stops working is agents: turning something loose on a real codebase and trusting what comes back. You already know that failure, because you have watched it. Output that is almost right, close enough to survive a quick look and wrong in a way that costs somebody a day. A review nobody can really do, because nobody can follow what changed. An afternoon spent debugging code that no one on your team wrote.

That is not a tooling problem and it is not a prompting problem. It is that nobody on the team has yet worked out what the agent is allowed to do, what has to pass before anything merges, and how you find out quickly when the output is confidently wrong.

I spent two years working that out, on a production system a real business depends on, because I had no choice.

What this actually is

I help you and your team build the tools that you want, using AI. I provide support and guidance, but it's your team on the tools and your team gaining the experience.

Your repository, your stack, your problem. We pick something real, an internal tool you've never justified building or a piece of your actual product, and your developers build it directing agents, with me alongside them. I review what they and the agents produce, catch the things that will hurt you in six months, and put the guardrails in place so the next one is safer and faster.

At the end you have two things: the thing, working. And a team that has done it once, properly, with someone experienced watching.

I call it a cobuild: you build it, I'm there while you do.

This is slower and more expensive than me just building it

I want to be straight about that, because you'd work it out on day three anyway.

If you can hand me a clean spec, I'll build it faster and better. Same money, better outcome, this once. For some companies that is the right call, and if it's the right call for you I'll say so on the phone.

But "hand me a spec" is carrying a lot of weight in that sentence. The things worth building tend to depend on what your team knows and has never had to write down: why that integration behaves the way it does, which edge cases are real and which are folklore, what the last two attempts got wrong and why. Getting that out of their heads and into a document I can build from is slow, lossy, and sometimes it is the entire job. Every "not quite, it's more like…" in a review is that transfer failing in front of you.

A cobuild skips the transfer. The people who already hold it build the thing, so what they know goes straight into the software instead of through a document and back out again.

This is for the other situation: you have a list of things your team has never built, and you'll still have that list next year. Paying a contractor to work through it one item at a time is not a plan. Paying once so your own people can work through it is.

You're buying the second thing, and the fifth. The first one is just where we do it.

This is for you if

This is not for you if

What five weeks looks like

First, a 30-minute call, free. You tell me what you'd build. I tell you whether this is the right way to get it or whether you should just hire someone. I've no interest in selling you five weeks of the wrong thing.

Before week 1, two days of mine. I read the codebase properly, work out where agents will help and where they will hurt, and prepare the guardrails so week 1 is spent building rather than orienting.

Week 1, one day, remote or on site. We go through the codebase and pick the shape of it. I set up the guardrails: what the agents are allowed to touch, what has to pass before anything merges, how you'll know when the output is wrong. Your team starts.

Weeks 2–4, one day each. Your developers build. I review what they and the agents produce, in your repository, against your standards. We fix the process where it's failing, not just the code. Most of the learning happens here and most of it is uncomfortable.

Week 5, one day. It ships or it doesn't, and either way we write down what your team now does differently. You get the guardrail configuration, the review checklist, and an honest note on what they shouldn't attempt yet.

Where it ends up running

Building it, deploying it and getting it hosted are all part of the job, not a loose end left hanging in week 5. You finish with it running in a place you control, and no ongoing requirement for me. If you still need me afterwards, it hasn't worked.

Usually that means your existing stack and your existing pipeline, because you have both. Where you don't, or where the thing doesn't fit them, we work out what does: your own infrastructure, a cloud account in your name, a dedicated machine, a container, a program that runs anywhere. That's a call conversation rather than something I can put on a page. There's no one size that fits everybody.

From £6,000 + VAT. That is eight days of my time. Two days preparing, five days building alongside your team, and a day of review and handover across the five weeks. Five of those days are in your team's diary. The rest are in mine.

Bigger teams or longer engagements cost more; we work that out on the call. Remote is the default and it works. I've been fully remote for years. On site if you'd rather, and I'm in West Yorkshire. Travel and accommodation costs are extra, and would be agreed up front.

Who you'd be working with

I'm Joseph Edmonds. I've been building and shipping commercial software for around 25 years, and I run this through LTS Commerce Ltd in Shipley, West Yorkshire.

I'm pro-AI with a lot of scepticism. The potential is genuinely remarkable and the danger is genuinely real, and harnessing it carefully, productively and safely is the hardest thing I've tried to do in this industry. Teaching that is the offer.

One thing to know before you book: I'll tell you if something's a bad idea, and I'll tell you if what you've got is bad. Some people find that annoying. It's usually the reason I was worth the money.

The obvious questions

"Can't we just work it out ourselves?"

You probably can, and for a while you should. It's free. The trouble arrives later: code that looks right and isn't, reviews rubber-stamped because nobody can follow what changed, and a growing pile nobody wants to maintain. If you haven't hit that yet, you don't need me yet.

"Isn't there free training from GitHub and Anthropic?"

There is, it's good, and you should do it. It's generic and self-paced and it won't look at your code. That's the difference, and it's the only difference I'm going to claim.

"Why is this more than a training day from a training company?"

Because it isn't a training day. A trainer delivers a syllabus somebody else wrote to a room of people. This is five days inside your repository with someone accountable for what your team ships at the end.

"What if my team hates it?"

Tell me in week two and we stop. I'd rather refund four days than spend a month somewhere I'm not wanted.

"Are you going to make our developers redundant?"

No, and if that's the hope, this is the wrong purchase. This makes the developers you have more useful. It doesn't replace them and I wouldn't sell it to you on that basis.

"What about our code and our data?"

NDA before anything, and I'll tell you honestly which tools send what to whom, including where I think the vendors' answers are vague.


Book a call

I'm exploring whether there's enough demand to run this properly. There's no start date yet and nothing is scheduled.

If it sounds like your problem, tell me about it below and I'll reply personally. Not a sequence, not an automation, me. If it isn't right for you I'll say so, and if you'd be better off with something free or someone cheaper I'll tell you that too.

Registering isn't a commitment to anything, and I'm not asking for money at this stage.

One line is plenty.

Optional.

One person reads these. Me. Usually same day.

How I handle your data. I use what you send to reply to you and to talk about the work. My basis is your consent, given by ticking the box; you can withdraw it any time by replying to any email. It is stored in a private Google spreadsheet only I can open, and Google process it on my behalf and nobody else sees it. No cookies, no analytics, no tracking of any kind on this page. Ask me to delete it and I will. The full detail is here, including how long I keep it and how to complain.

Questions, or you would rather just talk to a human? Email me at [email protected]. I answer everything myself.