Introduction

Here is the truth. Britain is behind. The United States is compounding advantage in artificial intelligence while our centre treats AI like theatre. AI is not a side topic. It is the topic. It changes how value is created, which jobs grow and which jobs decay and how our children should learn. Either we act with focus now or we get locked out of the compounding curve.

Whitehall and the missing response

The core problem is not a lack of speeches. It is a lack of a working centre. “If there is one word now to describe the Westminster regime, fake is the word. Fake meetings, fake decisions” (Dominic Cummings). He describes “fake scripted meetings” where “the conclusions are written by officials before the meeting ever happens” and says that is how modern government works in practice, not a parody of it (Dominic Cummings). He gives concrete proof of the culture. “We started monitoring sewage and provided real time data on disease spread. So the regime closed it down. We proved you could do vaccine research ten times more effectively. So they closed the vaccine task force. We created what I think is the West’s first data science and AI team inside a prime minister’s office” and the centre tried to vandalise it (Dominic Cummings).

On AI, the establishment posture is still jokes about chat bots. Dominic Cummings calls that “completely crazy” because the models are already matching and beating human performance across fields, and he warns the next few years will be fast and chaotic. The remedy he sets out is simple and hard. “Science and technology have to become embedded in the prime minister’s office as a core priority” with the authority to hire and buy at the speed of the world, not in thirty year epics (Dominic Cummings).

The American frontier operating system

The builders on the frontier have picked their method. They do not wait for one perfect model. They orchestrate many models and tools as one instrument. “When you use these models against one another the quality moves up quite dramatically. Our product gets dramatically better every time Gemini or OpenAI release a new model” (AI Now panel with Dave, AWG and Blitzy). They decided to “stand on the shoulders of giants” and route tasks across components rather than spend years trying to build a rival foundation model from scratch that would be obsolete on arrival (AI Now panel with Dave, AWG and Blitzy).

On software work they explain why old leaderboards are hitting ceilings. They pushed Swebench verified to eighty six point eight percent. At that level “the remaining questions are flawed” which means the bottleneck is the test not the method, so the next step is larger, more realistic tasks with extended inference time validation and rigorous review gates (AI Now panel with Dave, AWG and Blitzy).

The compute race and why energy is the new scarcity

Hardware scale is moving at frightening speed. “Elon went from zero to building Colossus one in one hundred and twenty two days. Everyone said it could not be done. He will not slow down. He wants to be number one” (AI insiders session discussing Elon Musk). This is the bitter lesson applied to the physical layer. More data, more compute, fewer artisanal tricks. “General algorithms plus huge data plus huge compute steamrolled the old approach” and it is steamrolling again at the infra layer (AI insiders session discussing Elon Musk).

Power becomes the constraint. “Colossus one has a self contained electricity source using natural gas cogeneration on site” so it is not waiting on slow grid timelines. Expect the biggest estates to colocate with their own power or the cheapest reliable sources they can secure, because the foot race is “very much winner takes all” and nobody wants the third best model when customers can see the difference (AI insiders session discussing Elon Musk).

The OpenAI insider view and the mission mask

“Secretive empire in the making, powered by billions, powered by one man’s ideology” is how the reporting framed it after a rare embed inside the lab. Senior figures struggled to answer basic questions about why billions must go to AGI rather than to solvable near term problems, a mission heavy nonprofit frame helped recruit talent, then the capital bottleneck forced a for profit structure and a tight embrace with scale compute (Karen Hao). The point for Britain is not gossip. It is to face the physics. Mission branding does not defeat scale economics. If you want capability you will need real capital, real compute and a real plan for safety and accountability at the same time (Karen Hao).

What this means for business models right now

Here is the business shift in one clean line. “No one cares about information. I can go to ChatGPT which summarises the world’s information. I can get it to be a lawyer. I can get it to be an accountant. I do not need information anymore” (Daniel Priestley). The move is from selling time to packaging your intellectual property as media, software, data and subscription. The leverage ladder is simple and ruthless. Time and skills at the bottom. Then intellectual property. Then media. Then data. Then software automation (Daniel Priestley). He also warns where the money sits. “Sixty percent of the money that can be spent is in the top ten percent of the people” which means you design for premium outcomes and scale, not for the bottom of the market (Daniel Priestley). The line that every British board should tape to the wall is the most brutal. “Nothing is safe from the emergence of AI. Nothing” (Daniel Priestley).

Why our schooling model is failing and what to copy instead

“The teacher in front of the classroom model of one person trying to educate many kids who are at wildly different levels just does not work” (MacKenzie Price). Alpha’s pattern is engines for mastery plus human guides who set standards, coach meta skills and protect motivation. The win is time. “We are giving kids back their most valuable resource which is time” so they can do real projects, entrepreneurship and life skills alongside mastery of core content (MacKenzie Price). Demand is real. “We have fifteen campuses opening across the country” with families able to plug in without losing their personalised flow if they move city (MacKenzie Price).

When MacKenzie Price sits with Daniel Schwartz, the Stanford dean, the watchwords for parents are culture as much as curriculum. Support not solve. Treat failure as fuel. Guide children to reflect and try again. That is how learners become resilient and independent in a world of fast feedback and fast change (Daniel Schwartz and MacKenzie Price).

The next three years if Britain does not act now

Put the threads together and project forward. The frontier will keep shipping weekly improvements that flow straight into products because orchestration harvests each new model without a rebuild. Software teams will shift more work to extended inference and review gates while old tests saturate. Compute will concentrate around firms that can finance and power the largest training and inference estates, with energy and location as gating product features. Routine white collar work will tilt toward tools. The winners inside firms will be problem shapers, system orchestrators, data stewards and customer educators. The losers will be those trained to be compliant rather than creative.

Dominic Cummings warns that the political class remains ignorant of science and technology and that is a “recipe for catastrophe” if left unchanged. He also points out a hard fact. Modern reasoning models will do tasks that “an entry level graduate student might do” in physics and other fields and they already show their work in public systems. The direction of travel is obvious to anyone paying attention and he calls the coming period “unbelievably chaotic” which history suggests is exactly what happens when general purpose technologies hit old institutions (Dominic Cummings).

What each UK business must do now

  • Run a ninety day AI audit across customer journeys and back office, score manual effort, cycle time, error rate and quality, then fix the few flows that move the dial first

  • Build a small in house orchestra, three to five builders who glue models and tools to your data and ship weekly with a visible change log and hard metrics

  • Stand on giants and route tasks across Gemini, OpenAI and Anthropic so your product improves with every new model release (AI Now panel with Dave, AWG and Blitzy)

  • Convert one service into a subscription outcome with a clear guarantee so you stop selling time and start selling results at premium levels (Daniel Priestley)

  • Start your data flywheel, define core entities, instrument key flows, capture labelled feedback and tie it to customer accounts rather than a vendor interface

  • Make quality a first class layer with review gates and extended inference checks and re run your best prompts and tools on every model upgrade to harvest free gains (AI Now panel with Dave, AWG and Blitzy)

  • Train for the roles you actually need, problem shapers, system orchestrators, data stewards and customer educators, and publish a skills matrix so people can move up the ladder (Daniel Priestley)

  • Treat energy and latency as product features, place workloads where power is cheap and reliable and reduce human in the loop delays where quality allows (AI insiders session discussing Elon Musk)

What each family can do now to put children ahead of the curve

  • Daily mastery blocks, short focused sprints in maths and reading with a simple dashboard for accuracy and time to mastery so progress is visible (MacKenzie Price)

  • Use routed lessons not copy and paste chat bots, keep the adult as guide and standard setter so the child learns how to learn not how to copy (MacKenzie Price)

  • Build an interest graph for your child and pair core skills with what they love so motivation sticks and content stays alive (MacKenzie Price)

  • Teach meta skills on purpose, plan work, debug calmly, read unclear specs, ask good questions and reflect honestly after mistakes (Daniel Schwartz and MacKenzie Price)

  • Replace passive evenings with real projects, three per term that end with a demo, a sale or a performance so reality is the teacher

  • If your school refuses to modernise, form a micro school pod and borrow the Alpha pattern of engines for mastery plus human guides and shared standards (MacKenzie Price)

A note on courage and controversy

It is not Brussels or a culture war that threatens Britain most in the next three years. It is fake control at the centre. Fake meetings. Fake responsibility. Fake decisions. The refusal to put science and technology at the core of the state. The frontier will not slow down for our rituals. The people who win will be those who face reality first and act. Britain can still choose to act.

Action plan in one page

For business leaders and owners:

  • Appoint a named orchestration owner for ninety days with authority to ship and a weekly demo cadence

  • Wire a minimal stack now, one coding copilot, one text model, one vision model, one speech model, a vector store, a workflow engine, a logger and a review interface that your team actually uses each day (AI Now panel with Dave, AWG and Blitzy)

  • Ship three production automations in ninety days, one revenue, one operations, one customer support, and track cycle time, cost per ticket, error rate and satisfaction before and after

  • Convert one service into a subscription outcome with a guarantee and design for the top ten percent of buyers where most of the money sits (Daniel Priestley)

  • Start the data programme, define entities, fix data quality monthly, and close the loop between human feedback and model updates

  • Tie bonuses to shipped value, the scoreboard is working software, faster service and happier customers

For teams and individual contributors:

  • Move up the leverage ladder from time and skills to intellectual property to media to data to software automation and publish a personal plan to climb it this year (Daniel Priestley)

  • Pair with AI daily on one hard task and keep a public build log that shows prompts, tools, failures and wins

  • Track your own factory metrics, cycle time, error rate and quality, and compete with yourself to improve them

For families and students:

  • Daily mastery blocks with visible dashboards that show accuracy and time to mastery so parents and children can see progress at a glance (MacKenzie Price)

  • Routed lessons with human coaching so the child learns how to learn rather than how to copy and the parent supports rather than solves (Daniel Schwartz and MacKenzie Price)

  • Three term projects that end in public delivery so children learn to plan, ship and present

  • Teach meta skills explicitly, plan, debug, read unclear specs, ask questions and reflect

  • If needed, form a pod and borrow the Alpha pattern so personalisation travels even if your local option refuses to modernise (MacKenzie Price)

YouTube references:

  • Daniel Priestley: How to 10X Your Business Without Working Harder
  • Dominic Cummings: is AI already in control?
  • “NOTHING Is Safe From AI!”: Your Plan To Become Irreplaceable In 2026
  • OpenAI Insider’s Dark Warning on AI & The Hypocrisy of CEO Sam Altman
  • AI Insiders Reveal Elon Musk’s Master Plan to Win AI w/ AWG & Dave Blundin | EP #192AI Now: Elon’s $1T Package, Apple’s $600B for Trump & How Small Startups Win w/ Dave, AWG & Blitzy
  • Learning in the Age of AI: Critical Insights from Stanford’s Graduate School of Education Dean
  • A Government Insider On Why Immigration Got Out Of Control | Dominic Cummings: What Can Be Done?