How to compete with AI

Summary: lean on judgement, taste and orchestration

Round one: FIGHT

The job market is super volatile right now, within and without a job, especially in junior roles where the output is increasingly indistinguishable. The looming threat is that AI becomes cheap enough to replace “good enough”. Since 2020, the cost of AI inference has dropped roughly 6,000x. The 2026 DeepSeek model is meaningfully smarter than the original GPT-4 and costs a fraction of what that used to.

So rather than asking “Will AI replace me?” the more appropriate one is “What happens when the thing you used to sell becomes almost free?”

Ruben Hassid’s Substack

Intelligence used to be a bottleneck

Every project used to have a bunch of human dependents and a roadmap with a timeline longer than a couple of minutes. The bottleneck wasn’t laziness — it was that producing decent intellectual output took real time from real people doing real work.

AI didn’t remove the need for that work. It made the first serious attempt almost free.

That’s a subtle but important distinction. Because once first attempts are cheap, the number of attempts explodes. Before AI, you test one headline. Now you test fifty. Before AI, you build one deck. Now you build four versions for four different buyer types. Before AI, you ask one analyst one question. Now you ask ten questions that would never have been worth anyone’s time.

This is Jevon’s Paradox applied to knowledge work: when something useful gets cheaper, total consumption of it goes up, not down. AI doesn’t mean less work. It means vastly more work gets attempted — and you’re expected to manage it.

Average dies first

AI is extremely good at average. Average ad copy, average sales email, average market analysis, average audience insights, average legal summary (using the word deliberately here).

Average used to be enough — because it signalled effort. It took time. Now, average takes twelve seconds and looks suspicious. “Did you just use AI for this?” is the most damaging question a client can ask you. It means they couldn’t see the human in it.

The moat that most knowledge workers had — sitting there, thinking, typing, knowing the format, assembling the pieces — has been largely dismantled. So the market is starting to ask the uncomfortable question: why pay a human for output that looks like this?

That question is clarifying and tells you exactly where to go next.

Judgment is what’s left — and it’s expensive

When intelligence gets cheap, judgment gets expensive. Everyone can generate options. Few people can pick. Everyone can draft. Few people can decide what should actually exist or get cut.

The valuable person now isn’t the one who produces the first version fastest. It’s the one who can look at twenty versions and say: this one, this angle, this will fail with our buyer, this sounds right but solves nothing, this should never be sent.

That’s taste. That’s sharp judgement. And taste is trained through doing the work — thousands of reps, bad calls, failed launches, clients who didn’t buy, projects that got killed. AI has read about those experiences. You lived them. That gap is real, and it’s getting more expensive, not less.

The practical implication: get your taste out of your head and into files AI can actually work with. Not generic prompts — context files. Your standards, your constraints, your audience, the things only you know. An about-me.md, examples of work you love and hate, your writing rules. Stop with one-time prompts. A well-built context folder is reusable memory. Feed AI the part it can't scrape from the internet and move on from average.

Stop selling hours

Time spent used to signal quality because production was expensive. Now production is cheap, so time is a weak argument. What isn’t cheap: understanding the real problem, knowing what to ignore, finding the missing risk, making the decision easier, shipping the version that actually worked.

Outcomes. Taste. Trust. Responsibility. Those stay expensive.

The value has shifted to the person who builds the system that makes questions unnecessary. The answer person — the one who drafts on request, researches on demand, summarises when asked — is the first to get replaced. The system person asks: why do we keep needing this? Then builds the template, the workflow, the trained AI setup that gets other people 80% of the answer without asking. You remove the bottlenecks.

Use AI where you already have taste

Most people use AI backwards. “I’m bad at this — AI will do it.” That’s a trap. If you can’t judge the output, you’ll approve garbage.

Use AI where you already have domain authority. If you’re strong at sales, use AI to draft twenty angles — you’ll immediately see which nineteen are weak. If you’re strong at writing, use it to generate raw material — you’ll catch the fake voice. If you’re strong at operations, use it to map workflows — you’ll spot the missing handoff. Your expertise becomes the steering wheel, not the passenger seat.

That’s ultimately what separates the people who stay ahead: not prompting skill, not which tool they use, but whether they can look at what AI produces and know — with real conviction, backed by real experience — whether it’s right.

That judgment is still yours. And you better believe it’s worth protecting.