22 October 2026

The seven pillars

For most of my working life I have thought about success as four things: hard work, luck, marketing and skill. Get enough of each and something good usually happens.

I was missing two. Time was the forgotten fifth, and capital the forgotten sixth. AI has now changed the weight of all six at once, and in doing so it has shown me a seventh that was always there but that I never named: distribution. By that I mean the reach you already have before you spend a penny on marketing. The people who listen to you. The list you could email tomorrow.

This started as the outline of a talk, and the argument is simple. AI did not make success easier. It moved the difficulty somewhere else, and most of us are still looking for it in the old place.

Seven pillars in three groups

Seven things is too many to hold in your head, so I group them, and the grouping is where the argument lives.

Resources you bring

Time. How long it takes to get from an idea to a shipped thing.
Hard work. The effort you personally put in.
Capital. The money it takes to start, build and keep going.

Capabilities you apply

Skill. Knowing what to build, and being able to build it.
Marketing. Getting the thing in front of people who do not know you.

Position you hold

Distribution. The audience that already trusts you.
Luck. Chance: timing, who notices, and which algorithm decides to like you this week. You cannot control the chance itself, but you can control how often you are in its way, and whether you notice when it turns up.

In one sentence: AI has collapsed the first group, commoditised the second, and left the third as the whole game.

Time and hard work used to be the same pillar measured two ways, because hard work was how you bought time. Capital was the other way to buy it, since you could hire people. AI has broken both links, which is why time has collapsed while hard work has only changed shape.

Fifty years in four columns

Here is how I think the weighting has moved for one person or a small business, at four points in time. By weighting I mean how much a pillar decides who wins, not how much it costs. Each column is the whole outcome, so a pillar can only grow by taking share from the others.

Fig. 1 · The rebalancing

Each pillar scored from one to five for how much it decided who won in each era, then drawn as its share of the column. The scores are my argued positions, not measurements.

The scores behind it, for anyone who wants to argue with a particular number:

Pillar	c. 1975	c. 2000	c. 2015	2026
Time	5	4	3	1
Hard work	5	4	3	3
Capital	5	4	3	2
Skill	4	5	4	3
Marketing	3	3	5	4
Distribution	3	2	4	5
Luck	2	3	4	5

Around 1975, before computers. Everything is manual, so time and graft dominate. Time decides a lot because it is unequal: whoever can find more of it, or buy it by hiring, gets there first. Capital is the gatekeeper: premises, stock and staff, or you do not start. Distribution is your reputation on the high street. Luck matters least, because markets are local and slow, and you mostly get what you earn.

Around 2000, the PC and the early web. Technical skill is the scarce thing. I started as a developer in 1999, and being able to build a website at all was an edge. Capital buys the team of specialists you need. Search engines briefly rent distribution to anyone who understands them, which is why it dips.

Around 2015, cloud, social and app stores. Frameworks and cloud hosting cut the cost of building, and paid acquisition eats the savings. Marketing becomes the whole game: funnels, growth hacking, paid social. Luck rises because platforms and algorithms now decide who gets seen, and distribution starts to mean followers and email lists rather than reputation.

2026. Build time and build cost are close to nothing, and everyone’s output is competent. Time has fallen furthest of all: everyone has plenty of it now, so it decides almost nothing. Skill narrows to taste, marketing craft is cheap, and capital moves to compute and attention. Luck is the largest single factor, and distribution is the one thing a model cannot generate for you.

The paradox of skill

The idea underneath all of this is Michael Mauboussin’s paradox of skill, from The Success Equation. When everyone in a field gets better and the gap between them narrows, luck decides a larger share of the outcome, not a smaller one. Absolute skill goes up, the differences shrink, and what is left of the gap is filled by chance.

His best example is baseball. Nobody has hit .400 over a season since Ted Williams in 1941, and not because hitters got worse. Everyone got good, so the outliers disappeared.

AI is the biggest narrowing of skill there has ever been. Anyone with a frontier model can now produce a competent app, landing page, piece of copy or video. The floor has risen to a height that used to be the ceiling.

Fig. 2 · The floor rose

A sketch of the shape the studies below point to, not a plot of their data. Being above average used to be an advantage. Now it is where everyone starts.

The first wave of studies found that shape again and again:

In Navigating the Jagged Technological Frontier, Harvard Business School and BCG gave 758 consultants GPT-4. On tasks inside what the model could do, they finished 12.2 per cent more tasks, 25.1 per cent faster, at higher quality. The below-average consultants improved by around 43 per cent, and the top performers by around 17.
In Generative AI at Work, Erik Brynjolfsson, Danielle Li and Lindsey Raymond followed customer support agents given an AI assistant. They resolved 14 per cent more issues an hour on average. The newest and least skilled gained about 34 per cent, and the most experienced gained almost nothing.
Shakked Noy and Whitney Zhang’s experiment in Science found the same in professional writing tasks: the biggest gains at the bottom, and a narrower spread between the best and worst writers.

Every one of them says AI lifts the bottom far more than the top. Competence is no longer something you can compete on. I wrote about what that feels like for the people on the receiving end in Getting AI-ready is a people problem.

The exceptions are the interesting part, though. In the same Harvard and BCG study, on one task chosen to sit just outside what the model could do well, consultants using AI were 19 percentage points less likely to get the right answer than those working without it. And when Nicholas Otis, Rembrand Koning and colleagues gave 640 small business owners in Kenya a GPT-4 mentor on WhatsApp, the strongest businesses gained around 15 per cent while the weakest did about 8 per cent worse. The advice was not what separated them. What they chose to act on was.

That is the same finding seen from the other side. AI lifts everyone’s work to competent, and what separates people afterwards is judgement: knowing when the machine is wrong, and which of its suggestions to follow.

Where each pillar went

Time: collapsed

A weekend now does what a quarter used to. I build software by walking and talking to AI, and arriving home with something that runs, which is what Driftcasting is about. Doing it most days is how I know this pillar has gone.

The part people miss is that if time is no longer scarce for you, it is no longer scarce for anyone you compete with. Speed stops being an advantage the moment everyone has it.

Hard work: moved, not gone

The effort has gone to two places. The first is volume: because each attempt is cheap, the work is now in making many more of them. The second is the part only a person can do: the sales calls, the conversations with customers, getting the thing shipped, and the dull persistence of turning up. AI has taken the hard work out of the middle and left it at both ends.

Fig. 3 · Where hard work went

Effort across five stages of making something. This is where the effort goes, not how much it decides who wins: the total is about the same, but the build, where effort used to set people apart, has almost emptied.

Capital: fell, then moved

Capital used to be the gatekeeper. First it was premises, stock and staff, then a team of specialists: designers, developers, artists, salespeople. You could not start without it. One person with AI can now do much of that team’s work, so the cost of building a product is close to nothing.

The money has not disappeared, though. It has moved to two places that behave in opposite ways. Compute gets cheaper every year: image and video generation, inference, a capable machine at home. Attention gets dearer every year, because everyone is competing for the same hours of the same people.

Fig. 4 · Where capital went

The direction of travel, not a plotted dataset. One line falls every year, and the other only rises.

So AI is pushing your money towards the one input whose price only goes up. There is a smaller warning in here too. Taking lots of shots, which I come to under luck, is cheap per shot, but shots add up. Fifty rounds of video generation at today’s prices is real money. The cost curve is on your side; it is not at zero yet.

Skill: narrowed to taste

Building is a commodity now. What is still scarce is judgement: what to build, what to kill, and what good looks like when everything is competent. Cal Newport’s argument in So Good They Can’t Ignore You still holds, but the rare and valuable skill has moved from making the thing to telling the good version from the merely competent one.

Marketing: split in two

The craft half of marketing (copy, creative, ad variants, search content, email sequences) is floor level now. Everyone’s marketing is good, and that is exactly what makes it harder. You cannot win on competence, so you have to win by being distinctive. Rory Sutherland’s Alchemy is the book for this: when rational, competent marketing is everywhere, the odd, human-shaped idea is the one people notice.

The other half, the part that decides whether anyone sees the good copy at all, has quietly split off into a pillar of its own.

Distribution: the moat

Albert-László Barabási puts it most cleanly in The Formula. Success follows performance, but when performance cannot be measured, success follows the network. When every product is competent, nobody can tell the performance apart, so the network wins. Someone with 50,000 people who trust them has something no amount of skill, work or AI can produce quickly, because trust builds at human speed, not machine speed.

For me the sharpest point in the whole argument is that an audience is stored capital. It is money you already spent, mostly as time and over years, that you do not have to spend again. With no audience you pay in cash. With one, you have prepaid.

Merch is the proof. Someone with a following can put a t-shirt design on a print-on-demand service and have sales the same afternoon, having spent almost nothing. Without the audience, the same test needs a marketing budget. Kevin Kelly made the case for a small, direct audience in 1,000 True Fans back in 2008. AI has moved it from a good idea to the main event.

It is the pillar that has changed least in absolute terms and most in relative ones.

Luck: bigger, and partly yours

The paradox of skill says luck’s share of the outcome goes up. That sounds like bad news, because luck is the part you cannot control. But luck has two halves. The chance events themselves are out of your hands: the timing, who happens to see the thing, what an algorithm decides to show. How often you put yourself in their way, and whether you notice when one goes your way, is not. As the chance share rises, being good at standing in its way matters more.

The first half you control is volume. If AI cuts the cost of an attempt by 90 per cent, you can make ten times as many attempts for the same effort.

Fig. 5 · Buying tickets

The chance of at least one hit, if each attempt has a 1 in 50 chance of being picked up and the odds per shot hold.

Nassim Nicholas Taleb’s case for optionality in Antifragile is the formal version: lots of small, cheap bets, each with a limited downside and an open upside. The catch is that everyone can afford more tickets now. When everyone takes fifty shots, each shot is worth less, so volume is the new minimum rather than an edge.

The second half is noticing, and that has not got any cheaper. The psychologist Richard Wiseman put adverts in newspapers asking for people who thought of themselves as unusually lucky or unlucky, and studied the ones who replied. In his best-known test he handed them a newspaper and asked them to count the photographs in it. On the second page, in huge type, a message told them to stop counting because there were 43. The people who called themselves unlucky tended to miss it. The lucky ones tended to see it.

It is an illustration rather than proof: the people chose themselves and described themselves, and as far as I know nobody else has repeated it. But it fits much firmer research on attention, which finds that anxiety narrows what you notice. The lucky people were not getting better dice rolls. They were relaxed enough to see what was in front of them.

I should be careful not to overclaim this. Robert H. Frank’s Success and Luck makes a strong case that successful people badly overrate the luck they made for themselves. The half you control is real, but it is the smaller half.

You cannot make a chance event happen. You can stand where it is likely to turn up, and you can be the one who notices when it does. Everyone can now afford to stand in more places, so that is no longer the edge. Noticing is, and so is having people who will tell you.

That is also how the last pillars fit together. Taste decides which shots are worth taking. Distribution means each one gets a first look from people who already trust you. Luck, the half you control, is noticing the one that catches and following it.

What I would do with it

Do more of taste and distribution. Pick the one thing that is yours, and spend the time you saved building on the people who will vouch for it.
Design for luck. Take lots of small, cheap shots in public, and pay attention to which one catches. The shots are cheap for everyone; the noticing is not.
Stop counting on time and capital. Building fast and cheaply belongs to everyone now, including the people you hoped would be your customers.
Watch out for marketing. It looks like the lever, and it is the most crowded room in the building.

Seven, not four

I ran on four pillars for years and forgot two. The seventh is the one I would tell anyone to start on today, because it is the only one AI cannot shortcut. You cannot prompt your way to people who trust you. You can only start sooner.

It is also why I think a team of one matters more now than it ever has. One person with taste, a small audience, the patience to take fifty shots and the attention to see which one catches has most of what used to need a company.