Vibe strategising: when AI shortcuts reach the business case

Most of the complaints about AI-generated content are aesthetic. The em dashes. The “it’s not X, it’s Y” construction. The reply on LinkedIn calling something a pivotal moment for the industry when it is a routine product update.

Chris O’Brien, CTO of Advania UK, finds all of that mildly irritating and basically harmless. What concerns him is the version nobody is complaining about, because it doesn’t look like slop at all. It looks like analysis. And it is already turning up in business cases.

Chris has been a Microsoft MVP for nineteen years and has spent the last few of them focused almost entirely on AI, building services around it for clients and leading a large part of Advania’s own transformation. He is not an AI sceptic, which is what gives his warning its weight. His position, as he put it in a LinkedIn post that prompted this episode, is pro AI and anti misuse.

A phase of unintentional misuse

Usage is ballooning across every role and every function, and Chris is entirely behind that. What he sees alongside it is people taking less ownership of the process, and becoming comfortable with AI in the driver’s seat rather than the human.

It shows up in pitch decks. In email communications. In vendor proposals, strategy documents, industry analysis and competitor analysis. His inbox, like most people’s, is full of targeted outreach that is obviously generated.

None of that is wrong in itself. The problem starts where a team uses AI, the result becomes their output, and the ownership simply isn’t there. In his assessment, that is already producing bad decisions and bad outcomes for organisations.

Vibe coding solved the easy half

The clearest illustration is what happened to software development. Large language models excel at language, code, syntax and structure, so it was no surprise that generation got dramatically faster and the effort collapsed.

What didn’t collapse is everything that makes software adoptable by a business. How does it run? Is it performant? Is it scalable? Is it secure? Will it respond well to different scenarios? Has it been tested? In many cases the answers are no, because vibe coding isn’t optimised for that yet. Platform engineering, DevOps and SecOps are harder problems and far more specific to context.

Chris expects that gap to close, and sees it as the obvious next focus area for AI-assisted development. But the underlying asymmetry, where generation is easy and assurance is hard, is now showing up well beyond code.

Vibe strategizing

As Chris puts it, we’re not just vibe coding software. We’re vibe strategizing, vibe roadmapping, vibe proposal generating and vibe RFP responding.

The public examples are accumulating. He points to consultancy work delivered to a government client that turned out to be full of AI hallucinations, leading to a partial refund. And to a published security analysis whose listed vulnerabilities, in many cases, were not real.

Neither of those is a story about AI being incompetent. They’re stories about validation, fact checking and analysis of the output not happening before the work went out with an organisation’s name on it.

The test: can you stand behind every bullet point?

Chris’s own method has two halves. The first is staying in control of the input. He is extensive and directive, explains the data and artefacts he’s providing and how they should be used, rather than throwing a load of files in and asking for something back. The second is going through the outputs with a fine comb.

What that produces is instructive. AI is genuinely better at generating a strategy deck than he is, and what comes back is more slides, more density, more points expressed. Which is also where the danger sits. Because he needs to own every last bullet point, tagline and framing message, he finds himself taking away as much as anything else.

The result is a deck he can stand behind under scrutiny, defensibly, in front of a board. And that, he suspects, is what others aren’t doing. He describes watching senior leaders and budget holders being convinced by a shiny deck that was largely generated from a single prompt.

Subtraction as a test

David Bowman picked that up as a tie back to this season’s theme of ownership as subtraction. If you genuinely understand your domain, you should be able to look at generated output and say what isn’t needed: the fluff, the extra points that don’t back up the narrative or the story.

Which turns cutting into a diagnostic. Being unable to remove anything suggests you’re forwarding the work rather than editing it.

You own the outcome

Jarbas Horst took the argument to its conclusion. Regardless of how you use AI, whether it augments your skills or does the entire task, you own the outcome. When it’s presented to someone, when it’s published somewhere, you are responsible for what you’ve put out there.

He also named the asymmetry that makes the excuse so convenient. When AI-assisted work lands well, nobody credits the tool. When it goes badly, suddenly the tool was driving.

Chris’s response was the sharpest line in the episode: using “that was AI” as an excuse should almost be a firing offence, because if you can’t stand behind every part of it, it shouldn’t be there.

The failure mode that actually matters

This is where the distinction between kinds of slop becomes useful.

Chris is relaxed about AI’s cosmetic tells. The stylistic constructions are irritating at worst. He’s equally relaxed about AI-assisted content in general where the thinking behind it is real. As David observed, you can know something was AI-assisted and still find a genuinely good point being made by a smart person. The two things aren’t mutually exclusive.

What is harmful is subtler. Opinion stated as fact. Weak attribution and weak data points, presented so convincingly, so polished and so well articulated that the brain intrinsically assumes the whole thing is robust from bottom to top.

Or a citation that exists but, when followed, turns out to lead nowhere authoritative. Chris’s comparison is pulling a citation from the Daily Mail rather than the New York Times or the Washington Post. There was a citation, and AI presented it to look strong.

There’s a self-fulfilling loop underneath that, too. Marketing teams with a good handle on their LinkedIn or Reddit strategy are following playbooks created by the AI tools themselves, which effectively say that the more authority signals you have out there, the more likely people are to find your product. As David added, what frequently sits at the end of such a link is research a vendor produced to stimulate their own sales. Not peer reviewed, not anything most people would accept as an indicative source, but wrapped in a shiny output it reads as evidence.

Catching that requires something close to a critical journalist, or a senior analyst who can x-ray a piece of content. Executives and boards, Chris argues, aren’t currently attuned to it. And it is making its way into business case building and organisational decision making.

His longer-range concern follows from that. Big organisational decisions, opening in a new country or starting a new product line, will have AI at the heart of all the thinking. Looking back in ten years to assess whether the quality of that decision making increased or decreased will be very hard to measure, and the answer matters at the level of organisations, industries and possibly economies.

Why counting agents proves nothing

The conversation also covers measurement, where Chris sees real immaturity. He’s spoken to clients who are required to report how many AI agents are running in the business, a metric you could game by spinning up a thousand inconsequential agents in ten minutes.

Token maxing and prompt volume have the same flaw. They reward usage rather than outcome. What eventually changed the conversation, interestingly, wasn’t a governance insight but economics: as vendors moved from per-seat pricing to consumption models and token costs ballooned, throwing AI at everything stopped being free.

The better approach, in his view, is to hold onto the fundamentals of the business. What makes this particular organisation, in this particular industry, go faster and work better? Every organisation should be able to assess its own progress against that, and it would be far more effective than counting tokens.

He’s careful not to blame the people setting these metrics. They’re often further from the technology and not thinking about the mechanics, which he considers entirely natural. More than that, he sees the underlying drive as positive: boards and CEOs are trying to agitate their organisations into doing better, and that instinct is sound even when the specific measure isn’t.

Crossing the streams

His proposed fix is convergence. Technical people are too far removed from what makes their organisation work effectively, and don’t really understand the boardroom conversation, which is natural enough given they aren’t in it. But it leaves them directing their efforts without much sense of the organisational drivers.

Meanwhile business leaders and managers would benefit enormously from building an AI agent or two and getting their hands dirty on what’s easy, what’s hard and what’s actually possible.

Both sides carving out time on the other’s ground would improve the decisions being made in the middle.

The cost of being the dissenting voice

David closes with a pointed question. Are we currently in a moment where it’s career limiting to be the only voice in the room saying we shouldn’t do this?

Chris’s answer is balanced. Anyone sitting at the meeting point of technology and organisational strategy will spot things others miss, and it’s incumbent on that role to point out danger, risk and what’s being missed, particularly where an AI strategy rests on a hypothesis that isn’t founded in the right data or evidence. You’ve got to make sure that pitfall is avoided.

But he also offers a decade’s worth of perspective on moving into leadership. Leaders often get there for a reason. They don’t get every decision right and they aren’t perfect strategists, but their gut feel, expertise and experience often operates at a higher level of elevation. And your own perspective may not be right either. Sometimes what’s on the table is a bet you wouldn’t have made yourself, and it’s still the right bet for the organisation.

Asked what decision he’d take if it were his, Chris pointed at UK organisations collectively. Everybody says they’re doing AI and most can point to something real, but plot the ambition and the willingness to take bets and it isn’t there. People are being ultra conservative, tinkering around the edges, not wanting to invest. He’s upfront about his own bias as a services provider. But he’d like to see more bullishness, more ambition and more bet taking, while still demanding robust evidence of return rather than doing AI for AI’s sake.

Listen to the full episode of Fresh Perspectives, and read the Employee Attention Recession report in full.

Watch the full episode here