For more than twenty-five years, I have watched technology cycles move from the “magic” phase to the “utility” phase. I remember the early days at Nokia when we thought the mobile phone was just a portable telephone, only to realize it was actually a remote control for life. I saw a similar shift at easyJet, where we transformed from a travel company with a website into a digital-first commerce giant. Today, we are hitting a similar pivot point where the “party” of brute-force scaling is winding down, and the industry is finally starting to sober up. Are you ready for the era of the “Small Model” advantage and the end of the chatbot as we know it?
The Death of the Search Referral and the Pivot to Action
According to the latest Reuters Institute Digital News Report, publishers are bracing for a massive 40% decline in search engine referrals. Why? Because the “middleman” is now consuming the value of the “producer.” When ChatGPT or Google Gemini provides a full summary, the user has no reason to click through. This is an existential crisis for any product strategy reliant on top-of-funnel traffic.
For Product Leaders, the solution isn’t to fight the agents; it’s to build them. But we must move beyond the “chat” interface. Moving from reactive boxes to autonomous executors is a structural change in execution. As highlighted by IBM’s recent forecasts, this is the year when multi-agent systems move into production. These aren’t just bots that talk; they are agents that do—negotiating refunds, booking travel, or managing energy grids without needing a human to click “run” at every step.
The Brittleness Paradox and Agentic Governance
The honeymoon phase of flashy proofs-of-concept is over. Boards are no longer satisfied with agents experimenting in safe playgrounds. This shift brings us to the “brittleness paradox”: the more efficient we make our automated agents, the more catastrophic a single failure becomes if the foundation isn’t clean. We are seeing a move toward Agentic Governance—where the risk is no longer that the AI fails, but that it succeeds in ways we cannot audit or explain.
Businesses that treat AI as a mere feature are finding themselves structurally fragile. Those treating it as infrastructure—with the same rigour as a payment gateway—are building resilience. This requires a fundamental shift in our Product Ways of Working. You cannot build autonomous agents with siloed, project-based teams. You need empowered, cross-functional squads that own the outcome, not just the output.
Small Models vs. The Rot Economy
We must be wary of “Enshittification,” a term coined by Cory Doctorow to describe the lifecycle of digital platforms that eventually cannibalize user value for corporate extraction. We see this in the rise of “data tolls,” where platforms like Salesforce face scrutiny over increasing connector fees. To insulate your product from this “Rot Economy,” the real winners are turning to Small Language Models (SLMs).
- Reduced Cost: Ownership of fine-tuned SLMs lowers inference costs compared to general-purpose giants.
- Privacy and Speed: Large-scale enterprises like AT&T are moving toward local, specialized models to ensure data sovereignty.
- Reliability: Smaller, specialized models are more predictable and easier to wrap in safety guardrails than “one-model-to-rule-them-all” approaches.
Infrastructure is the New Innovation
In my time at EDF Energy, making digital the lead customer channel wasn’t about the flashiest tech; it was about making the technology invisible and reliable. The “Great Filtration” is now occurring: labs throwing more GPUs at the problem are hitting diminishing returns, while those innovating on architecture—like World Labs—are focusing on “world models” that understand spatial reasoning and physics.
If you are a CEO or a CIO today, your priority isn’t finding the “smartest” model. It is finding the most predictable one. We are moving from the hype of the demo to the hard work of the deployment. Success requires building agent-compatible architectures—moving toward microservice-based agent structures and robust orchestration frameworks that can survive the transition from copilots to the silicon workforce.
Stop waiting for a smarter model to solve your product flaws. The next leap isn’t coming from a larger version of what we already have; it’s coming from how you integrate autonomy into your core operating model. Audit your current workflows today: are they designed for human sign-offs and legacy friction, or are they ready for an agent to take the wheel? The organizations that win won’t be those with the loudest AI, but those that make AI an invisible, reliable part of their service design.
Leave a Reply