Here’s what’s on our plate today:

• 🧪 Inside the quiet race to turn intelligence into performance.

• 📰 AI trade's "kindness of strangers" phase, the World Bank's AI ultimatum, and why developing economies may gain the most.

• 🛠️ Weekend To-Do: map a workflow, read the operating truths, audit your pilots.

• 🗳️ Poll: what separates the AI winners from everyone else?

Let’s dive in. No floaties needed…

The Laboratory

TL;DR

• Two races, not one: the visible contest over who builds the smartest model has a quieter twin: who builds the machinery that turns intelligence into output.

• Adoption without impact: 88% of organizations use AI in at least one function, yet two-thirds never scale past pilots and only 39% see any profit effect.

• Models became interchangeable: AI-native companies design for model swaps, shifting the advantage to workflow design, access to knowledge, and autonomy rules.

• A category is forming: Replit, valued near $9B, sells the pipework rather than the model, and grew from $10M to $100M annualized in six months.

• The stakes: Gartner expects over 40% of agentic projects to be canceled by the end of 2027, so this race can be lost as easily as the first one.

Inside the quiet race to turn intelligence into performance

In motorsports, there is a common understanding that when teams compete, the better driver brings home the trophy. The idea is true as far as it goes, but it obscures the sport's deeper structure, because drivers can only compete on equal footing when their machines are equally capable. Inside that detail lies evidence of a second race, one not decided by throttle input or the nerve to hold a corner at speed. It is the race to build the best possible engine, transmission, chassis, and steering, the systems that let a driver's talent reach the track.

The reason this overlooked race matters as much as the driver is simple: even the best driver, sitting on the most powerful engine ever built, cannot win if the transmission and steering cannot translate that power into performance.

In artificial intelligence, a similar playbook is now taking shape. For two years, the industry has run one highly visible race over who builds the most capable model, measured in benchmark scores, context windows (the amount of information a model can consider at once), and the falling cost of running them. Every major announcement from Silicon Valley to Beijing has been framed around that contest, and it made sense while capable models were rare. The quieter race is over the machinery around the model, the systems inside a company that decide whether all that horsepower ever reaches the road. The clearest sign that this second race has begun arrived in June, when McKinsey published research on 15 companies built around AI from the start, spread across continents, industries, and stages of growth.

The firm expected to hear 15 different stories, and instead found the companies converging, independently of one another, on the same operating habits. Convergence like that suggests a discipline taking shape rather than a run of lucky experiments, and the results read less like productivity gains than like different businesses. A four-person sustainability venture serves 20 enterprise clients, producing compliance reports in minutes that once took a law firm weeks. A marketplace went from 50 deals per adviser to 3k running at once, and it got there by building agent analysts to absorb the volume of work rather than by hiring 60 times the staff, leaving people for the conversations where trust is won. In the language of the racetrack, these are the first constructors: teams whose edge comes from the machine built around the engine rather than from the engine itself.

Everyone has the engine

What separates these constructors from ordinary success stories is the backdrop they run against: the rest of the corporate world uses the same tools and produces almost none of this. McKinsey's latest global survey found that 88% of organizations now use AI in at least one business function, up from 78% a year earlier. Yet roughly two-thirds have not scaled AI beyond isolated pilots, and only about 1% of companies consider themselves fully mature in their use of it. Money tells the same story: just 39% of organizations can trace any effect on profits to AI, and most of those put it at less than 5%.

Access to intelligence has become close to universal, while converting it into performance remains rare. That is the gap between a grid full of identical engines and the handful of teams that reach the podium.

When the engine stops deciding

The strange part of that gap is that the industry created it by succeeding. The visible race worked: frontier models kept improving, and competition multiplied until capable intelligence became close to interchangeable. The companies McKinsey studied treat it exactly that way, designing their systems so any model can be swapped out for another the moment something better or cheaper appears. When every team on the grid can buy the same engine, the engine stops deciding races, and the deciding factor moves to where motorsport always hides it: the transmission, the chassis, and the thousands of small mechanical choices that determine how much power actually reaches the track. Inside a company, that machinery has unglamorous names.

It is the way work is divided between people and agents, the records an agent can search when it needs an answer, the rules for what it may touch without a human signing off, and the habits that spread a working method from one team to the rest.

The market has already started betting on that shift, and a commercial category is forming around the machinery itself. Replit is one example: a coding platform valued at roughly $9B that aims its AI agent at sales staff, marketers, and small business owners rather than at engineers, running it on frontier models built by others and growing from $10M to $100M in annualized revenue in six months along the way. A company like that sells transmissions rather than engines, and its growth reflects how many buyers already believe the second race is the one that matters.

Building the transmission

Buying the machinery, though, turns out to be the easy half, because the hard half is rebuilding a company around it. McKinsey's research on the agentic organization describes one bank running an 'agent factory,' where 10 squads of agents handle know-your-customer checks (the identity vetting banks must run on new customers), and another using humans to oversee squads of agents modernizing its aging core systems, cutting the time and effort involved by up to half. The underlying design principle is that processes are rebuilt end-to-end around the agents, with people moved 'above the loop,' meaning they supervise the system and step in on judgment calls instead of sitting in on every step.

That is a different undertaking from bolting an assistant onto existing workflows, and it comes with dependencies most companies have never managed. An agent is only as good as what it can find, so the knowledge a company never wrote down becomes the ceiling on what its agents can do. Autonomy also has to be earned in each context, a lesson one company in the study learned when an email agent tried to accept contract terms that still needed negotiating, and a human caught the reply before it went out.

The second race has its own crashes

All of this construction is happening without any guarantee that it is being done well, and the sport's long history of failed constructors already has a corporate parallel on record. Gartner predicted in June 2025 that more than 40% of agentic AI projects will be canceled by the end of 2027, blaming escalating costs, unclear business value, and inadequate risk controls. The firm estimates that only about 130 of the thousands of vendors selling agentic products are building the real thing, with the rest engaged in what it calls 'agent washing,' relabeling ordinary chatbots and automation as agents. Those numbers hold two readings at once, and they point in opposite directions.

In the first, the second race is mostly hype, a fresh way to spend money on systems whose value nobody has defined. In the second, these failures are what the early laps of a constructor's race always look like, with most machines breaking down before anyone learns to build one that finishes. The evidence currently fits both, and nothing in the data yet separates them.

What the moment leaves open is the same question the racetrack poses: whether the trophy goes to the most powerful engine or to the team that builds the best machine around it. That question is no longer just about the AI industry, because the operating habits McKinsey documented were found in ventures of four people and in platforms spanning the globe, which means the blueprints are already circulating and easy to copy. Every company that licenses a model is now, whether it intends to be or not, a constructor. Which of them ends up on the podium will depend on a simple distinction: whether they treat intelligence as the prize itself, or as horsepower that still needs a transmission, a chassis, and a steering system built around it before it wins anything at all.

Weekend To-Do

• Map one workflow end to end: Pick a single process your team runs weekly and diagram every handoff; you'll spot where an agent could absorb volume instead of a new hire.

• Read McKinsey's "seven operating truths": Skim the AI-native operating habits research to see what the converging companies actually do differently.

• Run a pilot-to-scale audit: List every AI pilot your org has started and mark which ever reached production; the gap is the real story, and usually the real problem.

Friday Poll

Headlines You Actually Need

• AI trade's "kindness of strangers" phase: One CIO warns the AI trade has entered a fragile "kindness of strangers" phase, dependent on outside capital and sentiment holding steady.

• World Bank's AI ultimatum: The World Bank is urging developing countries to embrace AI or risk being left behind as the technology reshapes the global economy.

• Why developing economies may gain the most: The World Bank argues that developing economies have more to gain and less to lose from AI, given their lower exposure to automation and significant productivity headroom.

Meme Of The Day

The Toolkit

• Regie.ai: AI sales agent that researches prospects, writes personalized outbound at scale, and handles follow-ups so reps can focus on closing.

• Replit AI: A browser-based coding environment with an AI agent that builds, debugs, and deploys apps from prompts, with no local setup required.

• Sourcegraph: Code intelligence platform with Cody, an AI assistant that understands your entire codebase so it can answer questions and write code that actually fits.

Rate This Edition