Are we just a software factory?

No. And the reason is worth setting out, because we build with the same AI agents the factories do.

Someone asked me a version of this the week we launched. AI-native, they said. So are you really a product dressed up as a studio: point some agents at a brief, let them churn out the code, and hand over whatever comes back. No craft in it, no one who has understood our business, no one in the room with us. It is a fair worry. From the outside, plenty of this does look like a vending machine for code. So here is the straight answer, and the line that matters.

What people are really asking

The fear is not that we are a consultancy with a bit of AI bolted on. It is the reverse. The fear is that AI-native means automated to the point of absence. A service takes a prompt and returns a pile of code, priced like software, with the craft and the people stripped out.

That worry is earned. There are tools and vendors built exactly that way, and they are measured by throughput: lines written, features shipped, pull requests merged in a day. For a prototype or an internal tool, that can be a good afternoon’s work, and we use those tools ourselves. The model breaks when a real business is depending on the result.

Agents do the typing, people do the work

Agents do a large share of the work at LevelFive, and we are not precious about that. They write code, draft tests, wire up the plumbing, and they are fast. What they do not do is decide what is worth building, or understand why your business needs it, or carry the consequence when a call is wrong.

That part stays with senior people. It comes down to a line we use for the whole model: agent-led delivery, human-led judgment. The agents are tools inside the work. They are not the thing you are buying.

Do you just build to the brief?

A tool built to generate code starts from a specification and optimises for output against it. We start a step earlier, with the problem your business and your customers actually have, because the brief is often the first thing that turns out to be wrong.

So the early work is not typing. It is sitting with the people who own the problem and getting the decision right about what to build, before a line of it exists. Code written fast against the wrong brief is expensive waste delivered early.

We embed with your team

This is the part a product cannot do. Our people sit with your team, inside your business units, close enough to see how the work really happens and to change it as we go. Questions get answered in the room, not in a ticket queue. Decisions get made with the people who live with them. We are there to help move the business forward, with you, not to drop code at the door and leave.

That closeness is where craft shows up. Before anything goes live, a person checks what a demo never has to. Does it hold up when the inputs are wrong? Is one customer’s data kept clear of another’s? Does a failure reach our own tools before it reaches a customer? Can the next person change it safely a year on? Agents help us answer those faster than we could alone. They do not get to decide the answers are good enough.

What actually makes software hard?

The hardest parts of building software are not in the code. They are the human ones. People file them under soft skills, which badly undersells them.

Someone has to get a room of stakeholders who all want different things to agree on one of them. Someone has to read a team as it forms, storms and settles, and keep it moving until it performs. Someone has to catch conflict before it stalls the work, and scope drifting the wrong way while it is still cheap to fix. And someone has to tell a customer, with the evidence in hand, that the thing they asked for is the wrong thing to build.

None of that is typing, and no agent does it. It is the hard part of the job, and it decides whether the software succeeds.

Who owns the outcome?

There is a line from an IBM training slide in 1979 that has aged well. A computer can never be held accountable, therefore a computer must never make a management decision. Swap in today’s tools and it reads like it was written this year.

An agent can write the code. It cannot own whether the result was worth having. Here at LevelFive, that ownership sits with a named senior person, working to a fixed price for a defined piece of software. If we build it faster with agents, that is our reward for being good at the work. If a call turns out wrong, it is ours to fix, not yours to fund. You always have a person to hold to account, on your side of the table.

In that world more code is a liability, not an asset. We keep it to as little as the job needs, and grow the value the work creates instead.

Where that leaves you

If you want something running by the end of the week to test an idea, the prompt-and-build tools are a good use of an afternoon, and we will tell you so.

If you have customers, data, and a business riding on the result, you need more than code that demos well. You need people who understand the problem, sit with your team, and answer for the outcome. That is the work we do.

If it is the work you need, start a conversation →.