In this piece
How to insulate your business from the coming AI access crisis.
The fastest wave you have ever surfed
In the last week alone, Anthropic released Glasswing. There is talk of Mythos. Claude Agents launched. OpenAI hinted at something new. The capability jumps are not slowing down. If anything, they are accelerating.
If you are building with AI right now, or building a business around it, the pace is intoxicating. You can do things today that were impossible six months ago. The tools are extraordinary, the access is easy, and the cost is almost nothing.
Almost nothing. That is the part few people are talking about.
Intelligence withdrawal
Santiago (@svpino) put it bluntly this week: “Intelligence withdrawal will be brutal.”
The model tokens you are using right now are heavily subsidised. Anthropic, OpenAI and Google are burning cash to drive adoption. They are selling you frontier intelligence at a fraction of what it costs to produce, because they want growth first and margin later.
The subsidies are already disappearing. Pricing is moving up. Santiago predicts $500/month and $1,000/month plans, and he is probably right. The Anthropic-OpenClaw split made the business model pressure visible to everyone.
Think about what this means in practice. You have built workflows around cheap tokens. You have automated processes that assume near-free inference. You have started to rely on capability that someone else is paying for. When the price reflects the real cost, some of those workflows break. Some of those automations become uneconomical. Some of the access you take for granted simply goes away.
As one commenter put it: “Token access is already a class definer. Those who can afford hundreds of dollars of token usage a day are not playing in the same field.”
That gap is about to widen, and fast.
Distribution withdrawal
There is a second withdrawal happening at the same time, and it compounds the first. Levikov (@levikov) laid it out this week: “I know people making $100K/month with mid content and I know people making $3K/month with incredible content, and if you cannot figure out why, you are optimising the wrong variable.”
AI has raised the quality floor on content. Anyone can write well, edit well, produce video. Good content is table stakes now. The advantage is distribution, and distribution is something you either own or rent.
If your audience lives on someone else’s algorithm, you do not have an audience. You have a loan. And loans get called in.
Content compounds linearly. One more post equals one more post of reach. Stop posting and the machine stalls. Distribution compounds geometrically. One more subscriber on your email list makes every future send more valuable. One more member in your community increases social proof for the next. “McDonald’s doesn’t make better food than the local diner. McDonald’s has 40,000 locations.”
So you are facing a double exposure. The intelligence you are building on could get more expensive or gated. The distribution channels you are relying on could shift or shrink. If you have not insulated against either, you are more vulnerable than you think.
The Leanpreneur response
I’m not arguing for abandoning AI. The tools are too useful and they are here to stay. Build your relationship with AI the way you would with any powerful but unpredictable partner: clear boundaries, more than one option, and things of your own that nobody can take away.
That is what the Leanpreneur thesis has always been about. Business fundamentals first, technology second. AI is one lever among many. The operators who thrive through intelligence withdrawal will not be the ones with the most tokens. They will be the ones who built something real while the window was open.
Here are five principles for insulating your business, with practical moves you can make today.
1. Ground yourself in businesses that do not depreciate
The principle: Have exposure to business models that AI cannot disintermediate. Real-world services. Physical products. Skilled trades. Things that require presence, trust, and human judgement in ways that do not compress to an API call.
Why this matters: If 100% of your revenue depends on digital knowledge work that AI is getting better at every quarter, your margin is on a countdown. You do not need to abandon knowledge work, but you need some gravity in the real world.
What to do:
Get exposure to real-world business models, either by owning one, advising one, or being in their supply chain. Hospitality, building services, HVAC, food production, trades. These businesses are not going away and they are underserved by the people who understand technology well enough to help them.
If you are a consultant or operator, take on at least one client in a physical, tangible industry. You will learn things about operations, margins, and resilience that pure-digital businesses cannot teach you. And your income will not evaporate because a model got 10% better overnight.
2. Align with model provider ecosystems
The principle: If you cannot beat the model providers, become part of their enablement layer. The companies building frontier AI are not going to the wall. Being in their orbit is one of the safest positions you can take.
Why this matters: I spoke to a Microsoft partner this week. He is part of a special projects team that funds organisations to support adoption: reviews, training, implementation. These programmes exist because the model providers need enablers. They need people who can turn their technology into working processes inside real businesses. That need is structural.
What to do:
Join partner programmes. Anthropic has the Claude Partner Network. Microsoft has its partner ecosystem. Google has equivalent programmes for Gemini. These give you early access, credibility, and a commercial relationship with the platform itself.
Get certified in whatever is relevant to the stack you work with. The badge matters less than being in the room when the roadmap changes.
Position yourself as the person who helps businesses adopt the technology. The model providers do not want to sell directly to every SME. They want partners who can do that work. Be the partner.
3. Build sovereign capability
The principle: Own what you can. Run local models where they are good enough. Decrease your dependency on any single cloud provider for mission-critical workflows.
Why this matters: Right now, a local model running on decent hardware can handle 80% of routine business tasks. Summarisation, drafting, data extraction, classification. You do not need frontier intelligence for everything. The frontier models should be reserved for the 20% of work where the difference between good and extraordinary actually matters.
What to do:
Invest in hardware that can run open-weight models locally. A decent setup is not cheap, but it is a one-time cost versus an escalating subscription. Think of it as owning versus renting.
Build your workflows so the model layer is swappable. Do not hardwire your processes to a single provider’s API. Use abstraction layers. If Claude doubles in price tomorrow, you should be able to route your simpler tasks to a local model or a cheaper provider without rewriting everything.
Review your current AI usage. Map every workflow that depends on a cloud model. For each one, ask: what happens if this costs 5x more next month? If the answer is “we are in trouble,” that workflow needs a fallback plan.
4. Own your distribution and community
The principle: Build assets that compound geometrically and that nobody can take from you. Email lists. Communities. Direct relationships. Things that do not depend on an algorithm’s mood.
Why this matters: Levikov is right that content is now a commodity. AI can produce content at scale. What AI cannot produce is trust. A model cannot replicate a community where people know each other, share real problems and hold each other accountable. And a community compounds in ways that content never does.
What to do:
Start building an email list today if you do not have one. Not a vanity metric. A list of people who opted in because they find your perspective valuable. Every send to that list is distribution you own.
Build or join a community with real network effects. This is what the Leanpreneur Community is for me. Every new member makes it more useful for every existing member, and no algorithm change can take that away.
Stop renting your audience. If your entire reach depends on LinkedIn’s algorithm or X’s feed, you are one policy change away from irrelevance. Use those platforms to attract. Use owned channels to retain.
5. Master the fundamentals that do not depreciate
The principle: Sales. Follow-up. Showing up. Being in the room. These skills are gaining value because everyone else is hiding behind automation.
Why this matters: I spoke to the best recruiter I know this week. No AI. Just phone, email, and a basic CRM. Incredible at sales, diligent with follow-up, and if he gets in the room, he converts. That skill set will become more valuable than ever as the noise level rises and the human signal becomes rarer.
What to do:
Block time for old-fashioned outreach. Pick up the phone. Send the email that is clearly not AI-generated because it references something specific and human. Meet people in person where possible.
Get good at sales. Not the manipulative kind. The kind where you listen to what someone actually needs and then show them, clearly and confidently, that you can help. This skill is leverage that does not require a subscription.
Be relentless with follow-up. Most opportunities die because someone did not send the second email. This has always been true and AI has not changed it. If anything, the flood of automated outreach makes genuine follow-up stand out more.
Build something real while the window is open
We are in a rare moment. The tools are extraordinary and the access is cheap. That combination will not last. The subsidies will end, pricing will normalise, and the gap between those who can pay and those who can’t will widen.
Everyone will be using AI. What matters is whether you have built anything underneath it that holds up when the terms change.
So get some exposure to real-world business, align with the ecosystems, own as much of your compute and distribution as you can, and keep your sales skills sharp. All of those still work if AI gets more expensive. That is what separates operators from dependents.
Intelligence withdrawal is coming. The Leanpreneurs will be ready.
If you want to get ready alongside other lean operators, the Leanpreneur community is a small room of people doing exactly that.
