The future of work, 2035

The median knowledge job in 2035 is running a fleet — a marketing 'team' is one human and eleven agents with names, quirks, and performance reviews. You don't manage people; you manage a swarm, and your calendar is mostly the moments your fleet pings you to break a tie.

A whole class of work is just being the human willing to sign. Doctors, lawyers, and engineers increasingly get paid not to produce the work but to underwrite it — their name, license, and personal liability are the product, attached to output an agent generated in seconds.

Because agents converge on the statistically-average-good, companies hire humans whose only job is to catch the soulless-but-correct: the brand-flattening sentence, the technically-fine-but-dead design. They can't write the brief, but they can feel when something's off, and that feeling now has a salary.

A booming service trade exists purely to put a human back in the loop on purpose — handwritten onboarding, in-person closes, voice-only therapy with a verified pulse. The premium isn't quality; it's the scarce, certified fact that a person actually showed up.

The bottom rung snapped a decade ago, so most 28-year-olds never did the grunt work that builds judgment — and they know it. A nervous cohort of mid-careerists fakes seniority they never earned, while a few who apprenticed under the old system command absurd fees as the last people who actually know how things break.

One person now runs what used to need forty: a 'company' is a human, a credit line, and a roster of agents that does sales, ships product, and answers support overnight. They feel less like founders and more like conductors who are slightly afraid of the orchestra.

Every morning your fleet has already met without you. You wake to a three-line digest — what they did, what they disagreed on, the one decision they refused to make alone — and your first act of the day is approving or vetoing, coffee in hand.

Before any agent acts, a 'rails' layer prices the downside — what happens if this is wrong, who eats it, is it insured. High-blast-radius actions physically can't fire until a credentialed human's key co-signs. It's the seatbelt that made the autonomy street-legal.

Every artifact carries an invisible, cryptographic record of exactly how much was human, which agents touched it, and what was reviewed versus rubber-stamped. Hover over a contract, a diagnosis, a news story, and you see its 'human percentage' like a nutrition label.

Your agents' accumulated context — every preference, past decision, hard-won lesson about your business — is the real asset, and it's portable. When you switch jobs, you don't carry a resume; you carry a vault, and onboarding is plugging it in.

Because anything digital can be fabricated, the workday now hinges on hardware that proves the physical: signed sensor feeds, location-attested presence, biometric 'I am here and I am human' tokens. Trust migrated out of the screen and into things you can't fake from a server.

Most pro software in 2035 isn't for creating — it's for reviewing fast. Tools are built around the diff, the flag, the one-tap approve/reject, optimized so a human can adjudicate a hundred agent decisions an hour without their judgment going numb.

Companies stopped bragging about how many people they employ and started reporting agents-per-human as an efficiency metric. A 200-person company quietly runs 6,000 agents, and 'we're lean' now means the ratio, not the payroll.

Raw hours collapsed, but intensity spiked. The actual human workday shrank to a few hours of pure decision-making, each call heavier than a whole week used to be — and burnout shifted from overwork to the relentless weight of always being the one who decides.

When agents can do the task, the license to be liable for the task is the whole moat. Licensing bodies — medical boards, bar associations, engineering guilds — became some of the most powerful institutions of the decade, gatekeeping the right to sign.

With the bottom rung gone, the only way to build judgment is to shadow someone who has it. Paid human apprenticeships roared back as a luxury good — you pay to stand next to a master, because the agents can teach you the answer but not how to know when it's wrong.

Since your fleet works around the clock, real-time meetings nearly died — except as deliberate ceremony. Going somewhere in person now signals 'this matters enough to synchronize human bodies,' and offices became cathedrals you visit, not factories you occupy.

The market quietly repriced everything around one scarce input: a human who'll take the fall. Two identical outputs sell for wildly different prices depending on whether a real, namable, suable person stands behind one of them.

No one knows how to grow a senior person anymore, because the junior tasks that forged judgment are all automated. The whole economy is quietly aging out of its expertise, and nobody has solved how the next generation of masters gets made.

When work passes through ten agents and three vendors, blame evaporates. Catastrophes happen and the post-mortem finds a chain where every link points to the next — the central unsolved governance problem is making someone, anyone, actually answerable.

Humans were never built to adjudicate a hundred plausible-looking decisions an hour, and rubber-stamping creeps in unnoticed. The deadliest failures of 2035 aren't agents going rogue — they're tired humans approving things they didn't really read.

Everyone now looks expert because their fleet makes them sound expert, so hiring, dating, and trust all broke. Telling real judgment from a well-prompted performance of judgment is the defining social problem, and there's no clean test.

When your eleven agents and your colleague's eleven agents negotiate against each other at machine speed, things spiral — price wars, doom-loops, accidental collusion — before any human notices. Containing emergent multi-agent chaos inside a single company is a genuinely open problem.

A lot of people quietly grieve that they no longer make anything — they only judge things their agents made. The quiet crisis of 2035 isn't unemployment; it's the hollow feeling of being a permanent supervisor of work you never touch.

Insurance and a co-signing human-of-record for any AI-generated decision — we make agent output legally safe to ship.

Won because the first wave of agent-caused lawsuits made 'who pays if it's wrong' the only question buyers cared about; now embedded as the default liability rail in three enterprise agent platforms and quietly setting the actuarial price of AI risk for the whole market.

A reputation layer for humans — verified, slashable track records of the calls you've personally signed, so accountability is portable and priceable.

Became the de facto resume after employers realized years-of-experience meant nothing; people stake real money on their own judgment scores, and a high Vouch rating now out-earns any degree.

Matches early-career humans with retiring masters for paid, agent-augmented apprenticeships that rebuild the broken bottom rung.

Rode the panic that an entire industry was about to lose its last people who know how things actually break; backed by licensing guilds desperate to keep producing credentialed humans, and the only scalable answer anyone's found to the missing-generation problem.

A control tower that watches your whole agent fleet, catches doom-loops and accidental collusion in real time, and forces a human to the wheel before damage compounds.

Took off after a string of public fleet-fratricide blowups (a retailer's pricing agents raced each other to $0 overnight); now the standard 'mission control' layer, because running thirty agents blind finally scared everyone straight.