Is This the Decline of SaaS and the Rise of Vibe-Coded Replacements?

We ran our Manchester AI Roundtable on 30 September. Senior leaders in a room: law, financial services, engineering, IP advisory, supported housing, proptech. No slides. No agenda to sell anything. Just an open conversation about where organisations actually are with AI. This article is a reflection of what was discussed on the day.

The thing that struck me most wasn't a tool, a model, or a governance framework. It was how consistently the honest conversations kept pulling away from technology and landing on the same place. People.

The first admission changed the tone of the whole morning

A partner at a leading IP advisory firm described a process they'd recently automated. Mailbox triage. Working well. Everyone in the organisation called it AI.

It wasn't. It was pattern matching.

"No one else has asked the questions we're asking of vendors," they said. "That makes you wonder if you're being too thorough. But I'm not sure we are."

That tension, between moving carefully and moving at all, set the register for everything that followed. It started with something more fundamental than governance or procurement. A lot of what organisations are calling AI isn't AI. Conflating the two produces inflated expectations, failed pilots, and a scepticism that makes the real thing harder to adopt when it actually matters.

Nine out of ten engineers said they didn't use AI. Nine out of ten were.

That's from our own client work, shared with the room. We'd gone into an organisation where engineers had signed a policy stating they didn't use AI. The technical review told a different story.

Nobody was surprised.

This isn't a discipline problem. When people are under pressure and a tool makes their work easier, they use it. The organisations treating this as a compliance issue will keep losing. The ones treating it as a signal, asking what the usage tells them about where the real friction is, are the ones building something durable.

The right response isn't enforcement. It's creating a route so sanctioned that the unsanctioned one stops being worth the risk. That means psychological safety before it means policy documents.

The firms furthest ahead didn't roll out a tool. They built a culture.

One law firm in the room didn't choose a platform and then try to get people to use it. They ran a six-month pilot where teams compared tools in real work, shared feedback openly, and had direct conversations with vendors about what wasn't working. By the end, the feedback channel had grown so active they'd had to create a second one.

When they made a decision, people understood it. They'd been part of making it.

That's not a nice-to-have. That's the methodology. Every organisation in the room that had seen meaningful results had arrived there the same way. The failure pattern was consistent too: leadership decision, rollout, flat adoption, confusion.

People plus AI. Not AI minus people. We keep saying it because the evidence keeps backing it up.

"You might be able to identify when someone plays a piano piece wrong. That doesn't mean you can play it yourself."

That line came from a participant discussing what AI adoption is doing to junior professionals: the lawyers, analysts, and engineers learning their craft in organisations where AI is producing the first draft of everything.

Senior practitioners in the room use AI to interrogate their own thinking. Junior staff often use it to replace thinking they haven't yet developed. One firm now requires disclosure when AI has been used in supervised work. Not to penalise. To have the right conversation. Is this a knowledge gap or a judgment gap? They're different problems.

Nobody in the room had resolved it. But the firms taking it seriously are building frameworks that develop judgment alongside AI fluency. The ones that aren't are producing a generation of professionals who can spot a wrong answer but can't produce a right one.

That has a cost. It just takes ten years to show up on the balance sheet.

Data readiness isn't preparation for AI adoption.It is AI adoption.

Several leaders arrived at the roundtable wanting to talk about implementing AI and discovered, in the conversation, that the prior question hadn't been answered.

One participant had spent the better part of a year consolidating resident communications from WhatsApp, Teams, email and text into a single system. That wasn't a delay to the project. That was the project. The AI layer came after.

The version of this that stings: if your data is in a poor position, AI won't fix it. It will surface more of the problem, faster, in ways that are harder to ignore.

ROI is anecdotal. That's fine, until it isn't.

Casework from three days to three hours. Technical reports from three days to thirty minutes. A custom internal tool replacing a CRM package the organisation was using at ten percent of its capability.

The hard wins existed. But across the room, tracking was informal and most organisations were comfortable with that. For now.

The moment that changes is when AI spending becomes a line on a board agenda rather than a pilot budget. At that point, "people say it's making things easier" won't hold. The organisations defining what success looks like before they start, and building the measurement to match, will have something to show. The ones that didn't will be explaining why the numbers are hard to find.

Governance has moved on. Most organisations haven't noticed.

A year ago, AI governance in the room meant a policy document and a rule about ChatGPT. The more mature organisations had moved into something different: operational discipline. Regular usage reviews. Disclosure requirements baked into workflow. Client-specific data protocols. Active conversations with vendors about what happens to data at inference, not just what the terms of service say.

Several were asking a harder question: whether they wanted to own the full infrastructure entirely. AI sovereignty. Not the right answer for every organisation. But a question worth asking deliberately. The default is to keep using the commercial API, accept the terms, and trust the provider. Most organisations haven't thought through what that commits them to.

The gap is closing. In one direction.

The organisations that treated the last two years as a period for careful, deliberate adoption are compounding. The ones that waited for clarity, waited for the right moment, waited to see what competitors were doing: that window is narrowing.

That's not alarmist. It's what the room showed us when it stopped talking about technology and started talking honestly about what adoption actually takes.

The Curve helps mid-market organisations navigate technology strategy, AI adoption, and software delivery. If you want to understand your current position and build a proportionate approach, our AI Discovery service is the right starting point.