McKinsey dropped The AI Transformation Manifesto this month — Singla, Sukharevsky, Lamarre, Smaje and Levin, in collaboration with QuantumBlack. Twelve themes that separate companies truly rewired for AI from their peers. It's worth reading in full. It's also worth unpacking what's actually new and what's old framing in new paint, because the themes that matter for builders are different from the ones that make good conference slides.
The central claim is the one every founder operating in this space already believes, now validated by a large consulting firm: technology alone doesn't create advantage. Capabilities do. The early winners at AI are the same companies that won at cloud, at mobile, at web — because they had the underlying capability to absorb a new technology and put it to work at scale. The tools are broadly available. The advantage is how fast you can apply them to real problems in the parts of your business that actually matter.
Which brings us to the second theme, and arguably the most important: economic leverage points are your best focal points. Every business has a few — the bottlenecks where a 5% improvement creates outsized value. Freeport-McMoRan found it in process yield. Toyota found it in supply chain integration. Most organisations don't have a leverage-point strategy. They have a use-case inventory. That's the tell. Long lists of AI experiments are a symptom of a leadership team that hasn't done the hard work of figuring out where the gravity actually is.
The third theme puts numbers on the stakes. McKinsey studied 20 companies across industries that have proven themselves AI leaders: average 20% EBITDA uplift, one-to-two year break-even, $3 of incremental EBITDA for every $1 invested. These companies concentrated on one to three domains. They reinvented them — didn't improve them, reinvented them. Stage-gated investment, clear accountability for the business KPIs that mattered, and then they kept going. If your transformation plan is a series of incremental wins, you're not playing the same game.
Theme four is the one most executives quietly dread: the senior business leaders have to be tech- and AI-capable. Not IT leaders. The P&L owners, two or three levels below the CEO, who actually decide what gets built. McKinsey doesn't soften this — they don't have a single success story where senior business leaders weren't in the driver's seat. If your plan to "get more AI" is to hire a head of AI and point to them, you've already lost. You need business leaders who can conceptualise, build and run AI systems that power key parts of the business. That's a different hiring bar. It's also a different promotion bar.
Theme five is the people piece — what McKinsey calls the 30-70 shifts. More than 70% of tech talent should be in-house. More than 70% should be "doer" engineers who build great software-based solutions, not managers who coordinate it. More than 70% should be at competent or expert skill levels. Small, highly-skilled teams outperform large armies of lower-skilled staff. This is the structural case against body-shop consulting. The important line sits underneath that: as agents take on coordination and execution, human roles shift up the value stack. Engineers spend less time coding and more time designing architecture, workflows, constraints and quality controls. That shift is what the operating model has to absorb.
Theme six is the one we sign every engagement under: speed is the defining organisational advantage. The businesses that win don't have better technology — they redeploy resources faster, they empower teams to act without excessive dependencies, they reduce the latency between insight and decision and decision and action. McKinsey calls it the metabolic rate of the organisation. It's the right metaphor. Most enterprises are metabolically slow because coordination overhead has accumulated over decades. You don't fix that with an AI tool. You fix it with a platform, a clear operating model, and discipline about what teams own end-to-end.
Theme seven is where we nodded the hardest. Tech platforms are strategic assets. Invest in them that way. Platforms determine execution speed. Platforms drive unit cost down through reuse. Platforms put tech and data into the hands of the people who need them. Platforms are what allow AI to scale responsibly. Leading companies manage platforms strategically — dedicated teams, roadmaps, budgets, target service levels, users whose needs shape the evolution. Understanding your technical architecture is now as essential for a senior executive as knowing your P&L. If that sentence feels like an overreach, that's the exact gap McKinsey is naming.
Theme eight — productising data — is the adjacent twin. AI needs masses of high-quality data to be useful. In most organisations, data is still the constraint. Not because there isn't enough of it. Because nobody owns making it easy to consume. Scaling AI starts by treating data as a product — discoverable, accessible, consumable, with clear owners. Over time the game shifts to enrichment: deepening quality, context and uniqueness. Data is a business-owned performance asset, not a ticket queue for the analytics team.
Themes nine and ten are about what happens between the lab and the customer. Adoption and scale are where most AI programmes die. You predict equipment failures days in advance, and maintenance still runs on a calendar, because nobody rewired the downstream process. The AI works. The organisation around it didn't change. Scaling is a different challenge again — modular architectures, a well-choreographed dance between central teams and receiving units, run-cost discipline baked in up front. And hovering above both: trust. McKinsey's line is sharp — no trust, no right to deploy AI. That's a threshold, not a nice-to-have.
Theme eleven is the one builders are living right now: agentic engineering is the next capability to master. Foundation models are capable of sustained autonomous work over long periods. Leading companies are ingesting unstructured data, extending their AI platforms with agentic capabilities, automating guardrails, and codifying what works into a repeatable playbook. Not a research project. A repeatable playbook. This is exactly the ground-truth / skills / agents pattern we've been building on. What's new in April 2026 is that the McKinsey partners writing this expect it to be a mainstream enterprise capability by year-end. If that's right, the window for building the capability from first principles is closing fast.
Theme twelve closes the loop: learn, unlearn, relearn. The half-life of skills is shortening. The organisations that can compound learning outpace the ones that can't. CEOs can accelerate this by taking their leadership team on the journey themselves — until every C-suite member hits the conviction point where the strategy becomes obvious. That's not a slogan. It's the moment the transformation starts to self-propel instead of needing to be pushed.
Read the whole manifesto and a pattern emerges. Themes one through six are about conviction and focus. Themes seven and eight are about the foundations — platforms and data — that let the conviction scale. Themes nine, ten and eleven are about the mechanics: adopting, scaling, and doing agentic work responsibly. Theme twelve is the flywheel. The order matters. If you try to do eleven before seven, you end up with an agent factory without a factory floor. If you try to do three without four, you get ambition without ownership. If you try to do everything without one, you're just buying tools.
The Pareto reading is short. We've built the platform theme of the manifesto — secure, stable, scalable, AI-first — because that's what agentic engineering requires to be anything other than a demo. We've built the capability for agentic engineering — ground truth, skills, gates — because without that the platform becomes shelfware. And we've kept human judgment at the edges — Discovery and Validation — because that's the 20% where ownership and accountability actually live. Everything McKinsey describes as what leading companies do, we think of as the non-negotiables of how the work gets done.
The manifesto is useful because it pulls this conversation into boardrooms that were still debating whether AI was strategically important. That debate is over. The new debate is: are you building the capabilities, or are you still buying tools? The honest answer, for most organisations, is the second one. That's not a failure. It's the starting line.