SoftWeb Development Limited

02/10/2026

When I work on my own with AI tools, I can do everything: come up with the idea, build it, tidy it up. The only person I need to agree with is me.

Put five people on the same project and it gets much harder. AI means everyone is producing more, and all of it lands in the same place.

I do not think the answer is one developer per product. One person can only keep so much in their head, and it is risky if they leave.

What I think matters more is having someone whose job is to keep the whole thing joined up: deciding what not to build, and making sure five busy people are building one product rather than five.

30/09/2026

A user story is a short description of what a customer needs, and it has been the backbone of software teams for twenty years.

It was always meant to start a conversation, not replace one. The real detail came from people talking it through.

AI changes that. An AI tool cannot sit down with your product owner and ask what they really meant. Give it something vague and it fills the gaps with its own guesses.

So teams using AI well are writing that conversation down, in enough detail that there is little left to guess. At that point it stops being a user story and becomes a specification.

I think that shift will quietly change who does what on a lot of software teams.

28/09/2026

The person who created Claude Code recently shared five types of people he sees on his team: people who come up with new ideas, people who build them properly, people who tidy up, people who grow the product, and people who keep it running.

It is a good way to think about teams now that AI is blurring the old job titles.

What struck me is that AI helps some of those people far more than others. Turning a new idea into a working demo is now incredibly quick. Looking after a big, established system that customers rely on is still slow, careful work, because every change has to be checked before it goes live.

And if the idea people can now produce much more, the tidy-up and maintenance people have far more to deal with.

I wrote more about it here, including the role I think is missing from the list:

That page is not here. The services, work and writing are all still where they were.

04/05/2026

The dirty secret of the no-code revolution.

WordPress was sold on the promise that you wouldn't need engineers. There is a global ecosystem of agencies and freelancers whose entire business is making WordPress do what it was sold to do out of the box.

Salesforce was sold the same way. "Salesforce administrator" is now a six-figure profession.

HubSpot's partner directory lists thousands of agencies who exist to operate the platform.

A whole priesthood of consultants exists to operate the platforms that were sold on the basis you would not need them.

That has been a stable arrangement for fifteen years. AI is starting to break it, because agents are increasingly capable at the layer the consultants own.

I wrote about which categories of SaaS survive this and which do not:

Per-seat pricing is not the real threat. AI collapses the configuration layer that generic platforms charge for. Here is which SaaS categories survive.

29/04/2026

There's now software that listens to interview questions in real time, generates answers using AI, and displays them on a hidden screen overlay. Your screen sharing software cannot detect it - that's a design goal, not a bug.

The answer isn't trying to detect it. The answer is redesigning the interview so that it actually tests what you need to know.

I've written up what still works - including some simple shifts that hold up even as these tools improve. Link in the comments.

22/04/2026

The first week of a project rescue rarely starts with a plan. It starts with listening and understanding what has actually gone wrong, not what people think has gone wrong.

I wrote a day-by-day breakdown of what that process looks like and what we are working to understand before we change anything.

Link in comments.

14/04/2026

Most software projects do not fail suddenly. They fail slowly, with warning signs that get missed along the way.

I have written about the seven most common ones, and there is a free project health diagnostic on the site to help you assess where yours stands.

Link in comments.

12/04/2026

Only 22% of organisations are getting real value from their AI investments, according to Gartner's latest research.

If your organisation is in the 78%, the answer is unlikely to be a different AI tool.

The problem is not technical. It is structural:

Most organisations are designing the ceiling before the foundations are stable. They rush toward agent-based replacements of core systems while skipping the steps that have always mattered in enterprise technology.

They measure AI maturity by the number of tools deployed or pilots completed. Neither measures value. They measure motion.

And they treat data readiness as a future problem rather than a prerequisite, which means implementations proceed on assumptions that quietly undermine every output the system generates.

The organisations getting value did something different. They defined the business outcome before selecting the tool. They treated data quality as a blocker. And they put someone in a position of genuine authority to hold the organisation accountable for the gap between investment and return.

The gap between 22% and 78% is a governance gap. And unlike the technology, governance does not improve automatically with the next model release.

Read the full analysis here: