Make Hay, Then Reinvest

Every gold rush has the same problem. Most people sell shovels. A few sell shovels, and quietly buy the land.

We are in an AI gold rush. The proof landed in the last fortnight. Stripe ran Sessions 2026 with 288 launches, including a Machine Payments Protocol that lets agents pay other agents directly. Anthropic spun up a one and a half billion dollar joint venture with Blackstone, Hellman & Friedman and Goldman Sachs. OpenAI raised four billion from nineteen investors at a ten billion valuation, in a venture with PwC. Visa and Mastercard are now openly racing each other to become the rails for AI-initiated commerce. The wave is not coming. The wave is here.

Most commentary will tell you to ride it.

This is not that piece.

This is the piece that asks what the operators winning the hardest are actually doing with the money. Because, almost without exception, the people I see riding this wave well are doing two quiet things in private.

They are building distribution they own outright.

And they are reinvesting into businesses AI is unlikely to disrupt for a long time.

That is the actual play.

Sell the shovels. Buy the bedrock.

What riding the wave actually looks like

If you talk to enough people delivering AI training, AI consulting, or AI-first agency work right now, a pattern shows up. High revenue. Variable margin. A clear, almost universal acknowledgement that the price points clients are willing to pay today may not hold once the model layer commoditises further. Almost all of them are using the window deliberately. Make as much hay as the sun allows. Then channel the proceeds somewhere it compounds.

The two assets being built fastest, in this group, are an audience and an ownership stake. Audience and email list, podcast and community (IRL even better). The kind of distribution AI cannot delete from your hard drive. And, separately, equity or operating roles inside businesses that have been there for decades and look likely to be there for decades more.

The behaviour is consistent enough that I think it is worth naming it, because the dominant AI commentary right now does not. The dominant commentary tells you to lean in harder. Build more workflows. Automate more things. The quieter behaviour, from the people closer to the front of this, is to take some chips off the table while the wave is at flood, and put them into bedrock endeavours.

The three hedges

There are roughly three hedges I see being built in parallel.

The first is distribution you own. Your email list, your podcast, your community, your audience. AI changes the model layer. AI does not delete a fifty thousand person email list you spent five years building. Distribution compounds independently of which platform is up or down, which model is leading, or which workflow tool just raised at a four billion dollar valuation. It is the only asset that gets stronger every week you do nothing technical at all.

The second is equity in businesses AI is unlikely to disrupt for a long while. Bakeries. Trades. Install businesses. Infrastructure. Hand-made anything. Boring is the point. These have physical, relational, or regulatory moats, and a margin profile that does not collapse the day a competitor figures out how to use Claude or ChatGPT. The instinct is contrarian to the room you are probably standing in. While everyone around you is talking about agents, the smart move may be to put a meaningful share of this year’s earnings into a business with a dough hook in it.

The third is your own operating layer. Lean systems for yourself. Context, automations, repeatable workflows that give you the leverage of a team you do not have to manage. Not slop. Not tools-for-tools’ sake. The actual scaffolding that lets one person operate at the scale of ten without scaling the headcount. This is the personal-leverage hedge, and it compounds the other two.

Distribution. Bedrock. Operating layer. The three hedges.

The bakery test

I have been working closely with one bakery for years. It has been there for decades, and almost certainly will be there for decades more. The revenue line has grown materially during my time there. The labour share of revenue has stayed well below the industry norm. AI sits in the operating layer, but the customer cannot see it. The customer sees the croissant.

That is the test. The bakery passes it. The croissant cannot be drop-shipped from a model card. The brand cannot be deepfaked. The early morning mixing routine cannot be ported to a SaaS subscription. There will eventually be patisserie-making robots in cheaper conglomerate operations, and there are already factory bakeries selling on price.

But the artisan, place-based, relationally grounded version of that business has decades of moat in front of it.

The point is not that everyone should buy a bakery. The point is that the businesses that survive this best look like that one. Find the bakery shape in your industry. The thing AI cannot impersonate.

Long-term games with long-term people

There is a phrase Naval Ravikant uses that I keep coming back to. Long-term games with long-term people. Almost everything good in business comes from that combination. Almost nothing good comes from the opposite of it.

This is what I have always aspired to, and the move I see other operators in this position making too. Step away from any wholesale reliance on a single transactional revenue line. For me, that line has been recruitment for over a decade and it’s probably why I always felt dissonance with recruitment as a service. It’s viewed too transactionally. For someone else it might be agency hours, or training cohorts, or consulting day rates. The same transactional vibes. The line changes. The principle does not. If your income model is one transaction at a time with one buyer at a time, you are exposed.

The replacement is a portfolio of long-term partnerships, with people you respect and intend to be in business with for years. The shape of the partnership matters less than the duration and the people.

There is a spectrum of how involved you can be.

You can back a business with capital, take a stake, and stay quiet. You can advise, with equity or a small fee, and turn up to the strategic moments. You can be an operating partner, embedded part-time, owning specific outcomes inside the business. You can run a function as an embedded operator with a multi-year commitment. You can co-own, building in alongside the founder, and treat it like your own.

The shape can shift over time. What does not shift, in the operators I most respect, is the time horizon and the loyalty to the person on the other side of the table.

One vehicle that has emerged for me, generalist by design, is helping established businesses future-proof and survive. The exact involvement varies. The intent does not.

Pick people you would want to still be working with in ten or twenty years.

Find a shape that fits where you are now.

Let the shape evolve.

The agent-to-agent edge case

The thing nobody is fully ready for, and which is going to matter quickly, is agents talking to each other.

Stripe’s Machine Payments Protocol went live in March, and supports agent-to-agent payment coordination across both stablecoins and fiat in a single protocol. Visa rolled out Visa Intelligent Commerce, with integrations into Anthropic, OpenAI, and Microsoft, and predicts that millions of consumers will use AI agents to complete purchases by the 2026 holiday season. Mastercard launched Agent Pay, with verifiable intent tokens, and Fiserv has already integrated it into the merchant platform. There is a working stack of six commerce protocols already in the wild as of April: ACP, UCP, AP2, MCP, A2A, and Visa TAP.

The question every business owner should be sitting with this year is simple. Is your business ready to interface with an agent that is not yours? Is your agent ready to interface with someone else’s? Most businesses are neither. The ones that will be ready are the ones with clean systems, clean context, and a clean operating layer. Lean again.

The audit

Five questions for any operator or founder reading this. No answers required. Sit with them for the rest of the week.

1. What is your replicability score? If your role can be done by an agent in twenty four months, what is your angle, and where is your moat?
2. What distribution do you own outright? Audiences, lists, communities. What compounds without any further input from you?
3. What share of this year’s earnings is going into bedrock? Bedrock is whatever AI is unlikely to touch in the next decade. If the share is zero, that is data.
4. What is the state of your operating layer? Is it a moat, or is it the bloat? Audit your tool collection.
5. Are your systems agent-ready? Could another business’s agent interface with yours by the end of next year?

The Leanpreneur thesis has not changed. Business fundamentals first. Technology second. Lean is the model.

Make hay. Never skip the hay.

The window is open and you will not be forgiven for sitting it out. But notice what the operators winning quietly are doing with the money. Then do that too.

Distribution. Bedrock. Operating layer. Long-term games with long-term people. That is the play.

If this resonated, the Leanpreneur Letter goes out every Saturday and the community is where this kind of thinking gets sharpened with other operators in the same position.