A weekend Long Reads episode built around Every's new Thesis Statements project — 100 builders and thinkers writing short essays on the future of work after automation. NLW reads through the first batch and argues the tired 'will AI take our jobs' debate is finally giving way to the questions that actually matter: how work, teams, companies, and human advantage get rebuilt.
"Will AI take all our jobs" is the boring question. The real future of work is finally getting specific.
AI doesn't just take jobs — it changes the entire landscape of how we work: what individuals and teams can aspire to, how companies organize, which skills matter. Every's Thesis Statements project pulls the people living that future into the discourse, and a pattern emerges across the essays: more expert human work after automation, companies rebuilt as human-agent systems rather than software factories, and human advantage shifting toward wisdom, weirdness, unpredictability, and the heart.
The AI industry has wanted to have the job-apocalypse debate for far too long. The productive conversations are about how AI changes the entire landscape of work — what individuals and teams can aspire to, how companies should organize, and which skills to prioritize.
Alongside its first conference, Thesis, Every announced a project where 100 builders and thinkers write short essays on the future they see coming with AI. Dan Shipper's framing: a small group of humans knows what work after automation looks like because they live it every day — and their ideas are largely missing from mainstream discourse.
— Dan Shipper, CEO of Every. From his essay "After Automation, There Will Be More Human Work Than Ever": Every's nearly thirty-person team hasn't fired employees in favor of agents. Agents take over the stable, repeatable, well-framed layer — but someone has to point them at the right thing, judge the output, catch what's wrong, and turn results into real decisions.
Shipper's mechanism: AI trains on whatever expertise can be made explicit, which collapses the value of default model output. Demand for what's different is demand for human experts — even as we approach AGI, making expert work cheaper creates more situations where expert judgment is needed.
The current generation of models only knows about past work; humans are alive to a specific time, customer, code base, and conversation in a way the training corpus isn't. That aliveness isn't just fresher data — it's a continuous perspective with running wants, concerns, and a read on what matters.
Paul Millerd argues we flatten work down to activities with a job description and a paycheck — while the people declaring "work is solved" are drunk on work themselves, with optional jobs that already deliver dignity, challenge, and purpose. There is no "after automation" unless we radically expand our conception of work beyond job-shaped form.
NLW's case for the more-work camp goes beyond continued demand for human expertise: we've been playing in one tiny lit quarter of what turns out to be a monumental room. Humanity will race to discover, build, and create more — a radical expansion of what we can all do is the deeper reason there will be more human work than ever.
Noah Brier argues the insidious failure mode of agentic engineering isn't buggy code — it's agents building fundamentally misaligned features and systems. The hard problem is keeping humans, agents, and humans-with-agents building toward a single vision, which makes the right metaphor a software company, not a software factory.
NLW's read on Brier's essay: the enterprise world of AI users understands better than the startup world that technology without human and institutional systems around it is just another set of buzzwords. How and to what ends we build those systems is the real question.
Tina Ha's scenario: an agent sees a $30,000 CRM contract delivering $12,000 of value and cancels in milliseconds, no meeting, no negotiation. Agents are rational actors, not loyal ones — they neither see nor care that your software is easy to use and looks good. Winners will build headless, machine-to-machine architecture or own the can't-move-fast layers like banking and compliance.
Ha's conclusion: companies will compete less on having the best model and more on the systems connecting decisions to real-world outcomes — task routing, data access, workflow orchestration, rule enforcement. These businesses are toll roads: if you don't want to pay, building your own bridge could take years and hundreds of millions in compliance.
Former a16z partner Sumeet Singh calls the winners "post-skeuomorphic apps." Early mobile fell into the trap of replicating the physical world; the breakthroughs — Uber, DoorDash, Instacart — invented workflows only the phone made possible. AI is at the same inflection point: the winning question is "what work can we invent that only AI makes possible?"
NLW's framework: doing what you already do faster and cheaper is a reasonable start, but assuming this capability-unlocking technology will be constrained to old workflows fundamentally misimagines what's possible. Opportunity AI takes more iteration and experimentation — which is why prematurely ROI-ifying enterprise AI efforts biases teams toward the same old things, slightly cheaper.
NLW is skeptical in aggregate of startups whose only job is to watch what people do and train agents to do the same thing. Agents will almost inevitably work in agent-native ways suited to their strengths — which then requires another level of organizational integration to sync those new processes back with humans.
Tom Critchlow's railway analogy: fast trains forced standardized time; fast agents force a new coordination mechanism. Agents operate in seconds, teams meet weekly, finance plans quarterly — each inhabiting a different present. His proposal is "standard status": a continuously updated record of goals, decisions, permissions, and constraints shared by humans and agents alike.
NLW flags a theme he'll explore heavily this fall — the shift from individual agents to team agents, where coordination mechanisms and new systems designed around human-agent alignment become the substance of the conversation.
Obo CEO Nir Zikerman: LLMs excel at concrete, verifiable tasks like coding, so those jobs fade — but roles built on ambiguity, open-endedness, creativity, management, and human interaction thrive at unprecedented scale. In film, actors and screenwriters aren't going anywhere, while streamlined financing and logistics let artists take more creative risk than was ever economically feasible.
Joe Hudson argues that when a model can draft the brief or diagnose the anomaly in seconds, politely, there's no reason to keep paying the emotional tax of a difficult genius. Leverage shifts from what you can do to how you show up: emotional clarity, discernment, and connection can't be copy-pasted from a training corpus.
One of NLW's long-standing beefs with AI-industry future projections has been a historic unwillingness to get into specifics. Agree or not with each essay, this series digs into exactly the question his earlier episode "The New Jobs AI Will Create" tried to address — and that shift from vague pronouncements to nuanced exploration is the real progress.
Bethany Crystal runs a company solo that would have required five humans pre-AI — but the bigger effect is a return to playful, experimental work: bingo card apps for friends, hyper-personalized playlists, a choose-your-own-adventure blog. In a post-AGI world, the quirky niches that got you pushed around the playground are what keep you human.
Emily Vernon: now that tasteful-but-forgettable brands can be generated with a few prompts for free, the tyranny of good taste is more oppressive than ever. The brain registers predictable as forgettable, and AI's whole premise is being predictable — which makes unpredictable thinking, cultivated off-grid and away from recommendation algorithms, the human's place in the loop.
Sari Azout: machines took the heavy lifting and value moved to the mind; as AI amplifies intellectual output, value moves to judgment, intuition, taste, and self-knowledge. AI can tell us what is probable but not what is worth wanting — so treat every automated task as attention handed back, and reinvest it in the squishy skills rather than manufacturing more work.