Martyn Rhisiart Jones, Madrid, Friday 20th March 2026

March 20, 2026. Yes, that’s right, the calendar has finally caught up with the grift, hasn’t it? Thank you, thank you, for bothering to read my latest steaming pile of corporate word salad: “Choosing The Right AI In 2026 Is No Longer About Choosing The Right Model.” Because obviously, in 2026, choosing the right model would be far too simple, far too honest. No, no, we’ve evolved beyond that. We’re now in the rarefied realm of “capability profiles” and “orchestral conducting.” Please, hold your applause until I’ve finished flogging this dead horse made of buzzwords.

Over at LinkedIn and Forbes, yes, those august institutions where thought leadership goes to die a slow, well-branded death, I regularly churn out these profound insights on management and technology trends. If you want more of this exquisite torture, just click ‘Follow’. Go on, do it. Join my network. It’s basically a cult now, but with better emojis. Feel free to connect on Twitter, sorry, X, I forgot Elon rebranded it while we were all distracted by the latest parameter count, or Facebook, Instagram, my podcast where I interview other grifters about how we’re all going to be conductors of agentic orchestras, or YouTube, where the lighting is always suspiciously flattering. And don’t forget to subscribe to my other LinkedIn newsletters: AI & Future Tech Trends (subtext: buy my consulting), The Intelligence Revolution (subtext: I’m very intelligent), and The Future of Work (subtext: your job is toast, but look how elegantly I say it).

In recent years, by which I mean the last five minutes of hype cycle, the entire discourse around AI has been: which model’s got the biggest… parameters? Ooh, look at the context window! It’s massive! Bigger than yours! New ChatGPT drops, new Gemini drops, and suddenly everyone’s waving bar charts like they’re winning the lottery. “Superiority!” they scream. “We’ve surpassed human cognition!” Meanwhile, the model hallucinates that Napoleon was a TikTok influencer and confidently tells you the capital of Assyria is… wait for it… Assur-ton.

But today, today, my friends, those questions are irrelevant. Obsolete. Quaint, even. General-purpose LLMs? OpenAI, Claude, Gemini? They’ve all plateaued at roughly the same level of “good enough to write your passive-aggressive email to Karen in HR without anyone noticing it’s not you.” For everyday nonsense, they’re interchangeable. Like choosing between three brands of beige carpet.

But, BUT, for the serious business, the enterprise, the high-stakes, mission-critical, shareholder-value-maximising stuff? Oh ho ho. Performance varies dramatically. Dramatically! Like the difference between a man who can parallel park and a man who thinks parallel parking is a conspiracy by the deep state to control urban space.

So the question isn’t “which model’s best?” anymore. Perish the thought. The question is: “which models best align with our needs?” Align. Like horoscopes. Or chakras. Or finding the right therapist who won’t judge you for still using Clippy in 2026.

Forget devouring benchmarks. Benchmarks are for suckers. The real AI leaders, the visionaries, the conductors, understand capability profiles. Yes. Capability. Profiles. We assess models the way we assess humans now. Not on one stupid metric like IQ or annual sales. No. Multiple competencies. Analysis. Creativity. Communication. Decision-making. And, crucially, culture fit.

Culture fit! Because nothing says “cutting-edge innovation” like making sure your AI isn’t too woke for the boardroom, or too autistic for marketing. Some models do structured reasoning like a chartered accountant on amphetamines. Perfect for finance. Others autonomously create and execute action plans like a deranged PA who thinks they’re in a Jason Statham film. Great for workflows, terrible if you don’t want your expense claims rerouted via Pyongyang.

And then, oh, sweet Jesus, there are the creative ones. Rapid iteration of ideas. Brainstorming. The ones that sound like they’ve read too much Brian Eno and not enough health and safety regulations. Ideal for marketing. Or design. Or anywhere you want plausible deniability when the campaign tanks.

And don’t get me started on the specialists. Harvey, CoCounsel, Spellbook for lawyers, because nothing says justice like an AI that’s been fine-tuned on billable hours. Abridge or AWS Healthscribe for doctors, because who needs a Hippocratic Oath when you’ve got HIPAA-compliant hallucinations?

Profiling these models is now an essential skill. Essential! If you’re not profiling your AIs like you’re Tinder-swiping for a soulmate who won’t murder you in your sleep, you’re basically living in 2025. 

Pathetic.

Now, tasks, risks, outcomes. Let’s pretend this is deep. Define the task. How does it support critical operations? Triage thousands of support tickets? Risk-score transactions? Generate boardroom reports from KPIs that were probably made up in the first place? No single “best” AI. Obviously. You assess against the workflow. Speed? Pattern recognition? Deep reasoning with explainability, so you can blame the AI when the SEC comes knocking?

Risk! Low-stakes: creative ideation, prototyping. Let the wild ones loose. High-stakes: healthcare, legal. Don’t use the poet laureate of LLMs to diagnose cancer. That’s how you end up with “the tumour has excellent narrative arc but lacks stakes in the third act.”

Outcomes! Efficiency? Agentic autonomy. Accuracy? Guardrails and reasoning. Innovation? Diverse, unconventional, boundary-pushing ideas. The kind that makes VCs wet themselves and then quietly pivot when the prototype sets fire to the server farm.

So here’s the framework, the repeatable process, the secret sauce: stop taking the shiny new model and finding problems for it. Instead, a radical idea, look at what you actually need to do, decide how much error you can tolerate before the shareholders revolt, define success, then pick the tools that fit.

As your AI maturity grows, like cheese, you move from one blunt instrument to conducting an orchestra of specialised models and agentic systems. Marketing gets the multimodal dreamers. Finance gets the compliant paranoiacs. Legal gets the ones that cite case law like it’s holy scripture.

Secondary benefits? Less vendor lock-in. Resilience. No single point of failure. Unless the whole orchestra goes on strike because someone forgot to pay the licensing fees.

But most importantly, most profoundly, you get to think of yourself as the conductor of an agentic orchestra. Baton in hand. Tuxedo optional. Each model plays its part. A symphony of specialised bullshit, responsibly governed, optimised for hitting those beautiful, beautiful business goals.

Thank you. You’ve been marvellous. No, really. Follow me. Connect. Subscribe. It’s the only way to ensure you never miss another profound revelation about how we’re all just maestros now. Maestros of nothing very much at all. But with excellent capability profiles.