Context

NLW dedicates a weekend "big think" episode to Every's Thesis Statements project, in which 100 builders and thinkers write short essays about what work actually looks like after automation. He reads through and reacts to roughly ten of the first 25 statements, organized loosely into how individuals, teams, and companies will need to reorganize themselves, adding his own commentary and a two-bucket framework (efficiency AI versus opportunity AI) throughout. For a PM, this is less a single argument than a curated set of distinct, usable mental models for how to think about organizational design, product strategy, and personal skill development as AI absorbs more well-defined work.

The Big Idea

AI doesn't eliminate work so much as it commoditizes the explicit, well-documented residue of human expertise, which collapses the value of default, easily-automated output while sharply increasing the value of human judgment, coordination, and taste, the things that can't be made explicit enough to train a model on.

Every CEO Dan Shipper frames this as the foundational claim across the whole project: the more a task can be automated, the more it reveals adjacent, higher-order work that specifically requires a human being present, current, and accountable, meaning the total amount of valuable human work doesn't shrink as automation expands, it shifts and, several essayists argue, actually grows.

Key Insights

Automating the routine increases demand for genuine expertise

Dan Shipper's essay argues AI "commoditizes the residue of human expertise", whatever can be made explicit enough to train a model on, which collapses the value of default AI output and creates new demand for what's different: differentiated human judgment. His concrete example from running Every (a roughly 30-person company): even with heavy AI assistance in customer service, writing, and engineering, the team hasn't replaced people, because every complex task still benefits from a human and an AI working back and forth in the same workspace. His reasoning: current models only know about work that has already been done, while humans are "alive" to a specific, current moment, a live customer, codebase, or conversation, in a way a training corpus can't be.

Founders who just "AI-ify" existing workflows will lose

Investor Sumeet Singh (formerly of Andreessen Horowitz, now at WorldBuild) argues the businesses that fail in the AI era are the ones that take an existing workflow and layer AI onto it, rather than inventing a workflow that wasn't previously possible at all. He calls this the trap of "post-skeuomorphism": just as early mobile apps replicated physical-world metaphors (a trash can icon that looked like a real trash can) instead of asking what a phone uniquely enabled, many AI products are digitizing old processes instead of asking what only becomes possible now. His reference case: Uber didn't digitize a taxi dispatcher's desk, it asked what becomes possible when everyone has a phone that knows where they are.

Once AI agents are the buyer, boring infrastructure wins

A contributor identified in the episode as investor and writer Tina (the name did not transcribe reliably from the audio) argues that as AI agents increasingly make purchasing and vendor decisions on a company's behalf, they'll behave as unsentimental, rational actors: an agent evaluating a $30,000 CRM contract that's only delivering $12,000 in realized value will simply switch providers at 2am, with no relationship or loyalty involved. Her argument: this favors companies that own either headless, machine-to-machine infrastructure or regulated, hard-to-replicate compliance layers (banking, wire transfers, regulatory approval systems), since as model capability converges, competitive advantage shifts from having the smartest model to owning the systems that connect AI decisions to real-world outcomes reliably.

Organizations run on mismatched clocks

Writer Tom Critchlow argues that AI has introduced dramatically faster-paced work (agents operating in seconds) without a corresponding change in how organizations coordinate (teams meeting weekly, finance planning quarterly, leadership revisiting strategy annually), leaving each layer of a company effectively operating in a different version of the present. His proposed fix, which he calls "standard status," is a continuously updated, shared record of goals, decisions, permissions, and constraints that both humans and agents can reference, similar to how railroads once forced the standardization of time zones once trains made local time genuinely inadequate for coordination. He's careful to note that staying synchronized isn't the same as staying aligned, since rituals like team meetings and planning cycles also serve a human function (being heard, motivated) that a shared status feed alone might not replace.

Knowledge value is shifting toward judgment and wisdom

Two essayists converge on a related claim from different angles. Oboe CEO Nir Zicherman argues that jobs requiring ambiguity, open-endedness, and creativity will scale even as concrete, verifiable tasks (like coding) fade, because AI is structurally limited at the former even as it excels at the latter, citing filmmaking as an example: human actors performing human-written stories aren't going anywhere, even as AI streamlines the surrounding logistics (financing, casting, scheduling) that don't require that specific human creative judgment. Joe Hudson, founder of The Art of Accomplishment, frames this more starkly as "wisdom work" replacing "knowledge work": when a model can outperform a human expert across multiple fields simultaneously, the differentiator shifts from what you know to how you show up, judgment, emotional clarity, and the ability to read a room, since that's the part that "can't be rushed and can't be copy-pasted."

Mental Models & Frameworks

Efficiency AI versus opportunity AI

NLW's own framework, offered in response to Sumeet Singh's essay: efficiency AI does existing tasks faster, cheaper, or better (a perfectly reasonable place to start, and not something to dismiss), while opportunity AI asks what becomes achievable that wasn't possible before AI existed. He argues most organizations default toward efficiency AI because it's easier to measure and justify, but that prematurely demanding ROI from every AI initiative systematically biases a company toward "the same old thing, but cheaper" rather than toward discovering genuinely new opportunities, which requires protected space for slower, less immediately justifiable experimentation.

Software company, not software factory

Noah Brier argues the dominant metaphor for AI-native engineering, the software factory optimized for throughput and defect reduction, is wrong. He proposes instead that building with AI agents is closer to Andy Warhol's Factory than Henry Ford's: both are about output at scale, but Warhol's model was organized around keeping every piece of work aligned to a single creative vision, not just stamping out uniform units. In his framing, the hardest problem in AI-native engineering isn't code defects, it's agents building features or systems that are technically correct but strategically misaligned with the product's actual vision and architecture, which makes the job of holding that vision together (traditionally the CEO's job) more central, not less, as more of the literal building gets automated.

Trade-offs & Nuance

Broad, expanded human work versus the illusion that "work is ending"

Former strategy consultant Paul Millerd pushes back on Silicon Valley pronouncements that "work is solved" within a couple of years, arguing this claim only makes sense to people whose relationship to work is already optional (they can sabbatical or job-hop and still get dignity, challenge, and purpose from lives centered on work). He contrasts this with people for whom work isn't optional, and argues that the entire framing conflates "work" with "jobs" in a way that ignores the vast amount of unpaid, uncounted work (childcare, caregiving, home labor) that already fills people's lives regardless of automation. NLW doesn't fully resolve the tension between Millerd's broader definition of work and Shipper's narrower, job-focused one, but treats them as complementary rather than contradictory: both converge on the conclusion that "more work" is the likely outcome, just via different routes.

Common Mistakes

Mistake: treating "will AI take all the jobs" as the central question

NLW opens by explicitly rejecting the binary jobs-or-no-jobs framing that's dominated AI industry discourse, arguing it's a "boring conversation" that obscures the actually useful questions: how work changes at the individual, team, and company level, and what skills and organizational structures should be prioritized as a result. The broader lesson: a framing that forces a yes/no answer to a complex structural shift tends to produce worse analysis than framing the same shift as a set of specific, answerable design questions.

Mistake: over-ROI-ifying AI experimentation too early

Building on Sumeet Singh's efficiency-versus-opportunity distinction, NLW warns that enterprises which demand immediate, measurable return from every AI pilot will systematically end up only finding efficiency-AI use cases, since opportunity-AI work is inherently harder to justify before it's been discovered. The fix isn't abandoning ROI discipline entirely, it's deliberately reserving some experimentation budget that isn't held to the same immediate-payback bar.

Practical Application

Sort your AI roadmap into efficiency and opportunity buckets

Explicitly tag current and proposed AI initiatives as either efficiency AI (doing an existing task faster or cheaper) or opportunity AI (doing something not previously possible), and check whether your roadmap is lopsided toward the former. If it is, consider whether your evaluation process (requiring proven ROI before funding) is structurally biasing you away from discovering opportunity-AI work.

Ask what workflow you'd design if you started from zero

Before automating an existing process with AI, explicitly ask Sumeet Singh's question: what would this workflow look like if it were invented today, assuming AI capability, rather than adapted from how it worked before AI existed. Treat a project that only makes the old workflow faster as a starting point, not the end goal.

Build a shared status layer if agents and humans are operating on different clocks

If part of your organization is moving at agent speed (seconds) while planning and coordination still happen on human cadences (weekly, quarterly, annual), consider Tom Critchlow's "standard status" idea: a continuously updated, shared record of current goals, decisions, permissions, and constraints that both humans and agents can reference, rather than relying solely on periodic meetings to keep everyone in sync.

Evaluate your own product as if the buyer were an unsentimental agent

If your product is priced or evaluated in a way that depends partly on relationship, brand affinity, or a human being too busy to reconsider a vendor decision, stress-test what happens once that evaluation is made by a rational agent optimizing purely on realized value versus cost with no loyalty. Consider whether your differentiation would survive that shift, or whether it depends on frictions that AI agents are specifically designed to remove.

Questions to Consider

• Is our current AI roadmap dominated by projects that make an existing workflow cheaper or faster, and have we deliberately protected any budget or time for discovering work that wasn't possible before AI, the distinction NLW calls efficiency AI versus opportunity AI?
• If an AI agent, not a human, were deciding at 2am whether to keep paying for our product based purely on realized value versus its cost, would it renew, or does our value depend on human inattention or relationship loyalty that an agent wouldn't factor in?
• Are any of our organization's coordination rituals, like weekly meetings or quarterly planning, primarily serving a human alignment and morale function that a faster, shared status update couldn't replace, or are they genuinely still necessary for reasons beyond information transfer?

Bottom Line

The more explicit, well-documented work AI absorbs, the more it reveals a larger space of human work that specifically requires judgment, coordination, taste, and presence in the current moment, which means the practical task for individuals and organizations isn't preparing for less work, it's figuring out which of these newly valuable, harder-to-automate capabilities to build toward.

Concepts to Explore

Every's Thesis Statements project

Every, the media and software company led by CEO Dan Shipper, is publishing short essays from 100 builders and thinkers on what work looks like after automation, as a public record of ideas from people close to the frontier of AI adoption, connected to their upcoming Thesis conference in November. NLW frames the project's value as pushing AI discourse away from vague, industry-wide pronouncements and toward more specific, falsifiable claims about how work actually changes, useful reading for anyone trying to move past the binary "AI will/won't take jobs" framing in their own planning.