This piece is the third of a five-part series drawing on Juicebox Founder Joel Pember's time at SXSW London 2026.
Why bolting AI onto your existing business is the wrong move, and what the factory story tells us about where the real returns are.
When factories first electrified, the ones that simply swapped the steam engine for an electric motor and changed nothing else got almost no benefit. The gains came only when firms redesigned the entire factory around the new power source. That took years. In some cases, decades.
AI is at exactly this stage. We keep replacing one component and leaving the legacy system around it intact.
Andrew McAfee from MIT Sloan School of Business put it cleanly at SXSW London: work is interactional, not individual. Make one person dramatically more productive, and you don't speed up the organisation; you move the bottleneck to whoever sits downstream. A process built to handle a dozen changes a week jams when AI starts producing a thousand. The congestion just relocates: to product, to legal, to marketing, to finance, each running at its own sustainable pace. The firms seeing real returns are the ones rebuilding end-to-end, not just individual tasks.
There's a useful name for what that rebuilt end-to-end looks like. Brian O'Reilly, COO of Writer, described the shift from occasional AI triggers to always-on AI teammates: systems that hold standing access to a company's knowledge, goals, and processes and run continuously. The difference isn't incremental. An agentic trigger is an AI bolted onto a task. An always-on teammate is a permanent node in the organisation, with real context and awareness. Crossing that line is what actually captures the productivity that the J-curve keeps promising.
Underneath the always-on teammate is what people are starting to call the enterprise brain: a unified, queryable layer of the organisation's own knowledge: documents, data, and increasingly a map of how the business actually fits together. Brand, customers, financials, operational intelligence.
Here's why this matters more than the model you choose. The AI model is rented and commoditised. A faithful, structured map of your own unique organisation is neither, and it can't be copied. Whoever builds the better enterprise brain gives both their people and their agents a sharper place to think from. The model running on top becomes almost a minor detail.
That shift in architecture has a practical consequence that hasn't been talked about enough. The economics of getting thinking done are changing shape. Labour was always the variable cost of cognition, but part of it now turns into a metered compute line you can dial up or down. You begin, in effect, to hire in tokens. That changes not just what agencies charge clients but how any business budgets, staffs, and scales, and it points directly toward the question of how much AI you should actually be using.
The early orthodoxy was to use as much as possible, treating heavy token use as a marker of productivity and sophistication. The discipline is already running the other way, and for good reason: power, not chips, is fast becoming the binding constraint on the entire AI build-out. Pick the right-sized model for the job. Use cheaper or open-source options where the frontier model is overkill, which is most of the time. Expect this to harden into standard architecture: tiers of models, from frontier down to on-device, with each task routed to the lowest tier that can handle it. Frugality is cheaper, more resilient, and far more energy efficient. It's not a virtue. It's a technical architecture principle.
The practical instruction is simple to state and hard to follow: find an end-to-end process you can rebuild around AI, instrument it so you can see what's working, and go. Not faster tools. A redesigned factory.
That's the structural argument. But rebuilding the factory only matters if what you're building around is worth protecting. Which is what Part 4 is about.
Thanks for reading, stay tuned for part 4, 'The biggest threat to your brand right now isn't a competitor,' dropping soon.
About SXSW London
SXSW London is the leading European platform for convergence and cross-industry collaboration, celebrating the intersection of business, technology, creativity, and culture. The 2026 edition ran from 1 to 6 June across dozens of venues, galleries, clubs, and independent spaces in Shoreditch, east London. Welcoming over 20,500 delegates from 77 countries in its inaugural year, it builds on the SXSW brand's decades of success in Austin and Sydney and its established reputation as a global platform for discovery.
About the author, Joel Pember
Joel Pember is the founder of Juicebox, a Perth-based design intelligence studio operating across Australia and Southeast Asia. Currently on sabbatical in Europe, Joel is spending time immersed in the ideas, conversations, and events shaping the next chapter of the industry. This series draws on his time at SXSW London 2026.
