Why agentic AI fell short of the hype

Welcome to this week’s edition. If you’re reading this with a coffee in hand and one eye on the rest of the long weekend, you’re in good company. I’ve pulled together a few pieces that felt worth pausing for this week – the kind that tend to sit in the back of your mind and reward further investigation.

Enjoy!

Why the ‘year of the AI agent’ never happened

Top tech writer Cal Newport has outlined one of the better reality checks on AI hype I’ve read in a while. His argument is not that agentic AI was a silly idea, but that the industry confused impressive demos with real-world readiness. Coding agents did get better fast, because large language models work well in text-based environments with clear commands. But the promised leap into everyday digital labour – booking, browsing, comparing, navigating, deciding – turned out to be much harder once those systems had to deal with clunky interfaces, multi-step reasoning and the small mistakes that can send an entire task off the rails. The bigger point for anyone offering a tech service is that when a product works beautifully in the lab but awkwardly in the wild, adoption usually takes much longer than the people selling it would like. The New Yorker has the story (paywalled so use Archive to access).

Jensen Huang wants to talk about factories, not chips

A smart summary of Jensen Huang’s latest Lex Fridman podcast appearance gets at the scale of Nvidia’s ambition. Huang’s argument is that AI is creating a new industrial layer, with data centres acting as “intelligence factories” and tokens as economic output. That matters because it shifts the frame from faster chips to a much bigger systems play involving compute, software, networking, cooling and power. The piece is a little ‘breathless tech acolyte’ in places, but it is still a useful guide to how one of the world’s most important executives thinks about demand, bottlenecks and why the real prize is owning more of the stack. Find out more here.

How Revolut built a growth engine that keeps compounding

A deep dive from a16z into Revolut’s 2025 annual report is worth reading because it gets beyond the usual fintech story of customer growth and asks what is really driving the machine. The answer, broadly, is not one killer feature but a compounding loop: more users, more products, more revenue streams, more primary-account behaviour and more operating leverage. Revolut grew revenue 46 per cent to £4.5bn in 2025, lifted profit before tax 57 per cent to £1.7bn, and now has a far more diversified business than the “cheap FX card” story many people still have in their heads. The piece is bullish, as you would expect from an investor, but it does a good job of showing how a fintech stops being a fast-growing app and starts looking like a serious financial platform. Check it out here.

LinkedIn is on a bot purge

Former Google Head of B2B Strategy Sarah Aird-Mash has an interesting theory about what LinkedIn is doing in 2026: cleaning out some of the spam and giving more exposure to people who are genuine subject matter experts. Her argument is that the platform is starting to favour relevance, expertise and posts with engaging insights, rather than endless churn of obvious AI filler. If she’s right – and many of the message boards on Reddit would agree – it’s good news for anyone building a reputation properly – and less good for the armies of bland, AI-assisted nonsense merchants. Read Sarah’s take here.

Why AI sounds like a mid-tier management consultant

A new Harvard Business Review report gets at a more interesting AI risk than the usual concern about hallucinations. The authors tested leading language models on classic strategic trade-offs and found that they kept gravitating towards the answers modern management culture likes best – differentiation, collaboration, decentralisation, long-termism – whether the context really justified them or not. In other words, the models often sound strategic while serving up the corporate equivalent of groupthink. That is worth paying attention to if you are using AI to pressure-test decisions, because the danger may not be nonsense. It may be advice that sounds polished and completely sensible right up until it leads you into the same mistakes that everyone else is currently making. Read the report here.

Could AI startups end up replacing service firms?

According to Sequoia’s Julien Bek, “the next $1tn company will be a software company masquerading as a services firm.” He suggests that some of the most valuable AI companies may not stop at selling tools to professionals. They may go further and sell the job itself – closing the books, handling claims, drafting contracts or taking on other chunks of work that companies currently pay service firms to do. His distinction in the piece between “copilots” and “autopilots” is a useful one, raising the question that if better models keep squeezing software into features, does more of the value move to whoever owns the outcome? Read Bek's take here.

Blagging it: report finds many marketers can’t do basic marketing

Cantankerous marketing guru Mark Ritson has a knack for making people bristle, but his latest piece lands particularly well because it comes with data. Drawing on Ipsos research, he says that a surprising number of American marketers do not understand some of the most basic concepts in the discipline – from positioning and penetration to research methods and the four Ps – and that the clearest dividing line is formal training. That will irritate the self-taught purists, but it is hard to dismiss the broader point: before marketers obsess over AI, purpose and boardroom influence, many of them may need to get a firmer grip on marketing itself. Read it here.

Why search captures demand others created

Rand Fishkin, the Moz founder and now head of SparkToro, has published a great piece on one of digital marketing’s oldest bad habits: giving search credit for demand it did not really create. Using Similarweb data across the 5,000 most-visited websites, he shows just how fragmented influence still is online – across social, news, email, content, entertainment and the rest of the web people move through before they ever type a query into Google. Additionally, while everyone obsesses over ChatGPT and AI search, Fishkin’s data suggests that Google’s use of AI inside search may matter far more than the standalone AI tools getting all the headlines. Find out more here.

AI prompt of the week: brand identity on a budget

Early-stage brands often waste time and cash chasing logos or websites before they’ve answered the basics. You don’t need a £50k agency – just clarity about who you serve, what sets you apart and how you sound different from everyone else in the market.

Help me define my startup’s brand identity and tone of voice. My context: [company type], [product/service], [target customer], [main competitors], [company stage], [3 words you want people to associate with your brand].

Create:

• 3–4 brand pillars that capture what your company fundamentally stands for (not vague adjectives like "innovative").
• 2 customer personas (primary buyer and key decision influencer) with their daily pains, goals and how they consume information.
• Tone of voice guidelines with 3 distinct modes (e.g. marketing, support, social) and 5 example sentences each.
• One-line differentiator that works across pitch deck, homepage and elevator conversations.
• Starter copy examples for homepage hero, About page, email footer and a customer case study headline.

Base this on how successful early-stage companies (like Stripe or Monzo) established memorable identities without big budgets or long discovery processes.

The hardest shift in scaling is often personal

A lot of founders know how to start something. Fewer know when their own habits have become the bottleneck. The challenge is not losing the edge that got the business moving, but knowing when instinct, speed and improvisation need backing up with better operating discipline. That switch is where many companies either grow up or stall.

Drop me a line

That’s all for now – thanks for reading. If you’ve got a better take on some of the points raised or even some mildly chaotic half-formed thoughts you want to get off your chest, then please do get in touch. Reading newsletter responses is always one of the highlights of my week.

Cheers!
Adam

​
​Unsubscribe · LinkedIn​