Skip to main content

Build What Big AI Won't

Frontier labs are coming for every niche that fits inside their chat interface. The work they structurally cannot do is the work that is opinionated. Build opinionated.

The Principle Big AI Is Applying​

Peter Levels (@levelsio) named it clearly in April 2026:

"BigTech will eventually come for all apps / startups / companies because they can fill the niches now that before could not because they were too small. [...] End game for their survival is simply trying to take every business. It's just capitalism. [...] They won't come at it in how you expect. ChatGPT isn't going to launch Remote OK or Photo AI or MAKE book. But they have and will make it easy in the chat interface to find a remote job, take AI photos, and learn how to build a startup."

Two forces are driving this:

  1. Marginal cost of filling a niche has collapsed. The old economic floor that protected small SaaS (too small for Big Tech to bother) is gone. AI lets frontier labs absorb the $100K-to-$100M-per-year niches that used to belong to independent operators.
  2. Competitive logic forces absorption. If one lab leaves a category alone, another will take it. Nobody is choosing to be the one that leaves money on the table.

Recent evidence: Anthropic reportedly testing a Lovable-style full-stack app builder. The pattern is not subtle and it will continue.

What Big AI Labs Structurally Cannot Do​

The deepest constraint on frontier labs is structural. Being opinionated is not the job of a frontier AI lab.

Their business depends on serving the widest possible user base. That structurally forces:

  • Neutral defaults. Anything that would alienate a meaningful user segment is off the table.
  • Mean aesthetic. The product has to feel acceptable to everyone, which means it cannot feel like anything specific to anyone.
  • Risk aversion. They cannot ship products whose core value is a contested take, a cultural stance, or a particular worldview. The lawsuit and regulatory exposure alone rules it out.
  • Scale-first optimization. Anything that is better at size 50 than at size 50 million is the wrong product for them to build.

This is the real line. Opinionated products, opinionated services, and opinionated communities are safe from absorption because a frontier lab absorbing them would dilute the very thing that makes them work. Labs cannot match them without becoming something they are not.

Every category of "unabsorbable" work listed below is a consequence of this single principle.

The Absorbable Zone​

Things frontier labs will absorb, eagerly, on timelines faster than you expect:

  • Generic productivity tools (note-takers, summarizers, drafting assistants, scheduling helpers)
  • Undifferentiated SaaS whose core value is "this task but with an LLM wrapper"
  • Creative generation in the default aesthetic (image generation, stock-style copy, mid-market video)
  • Data pipelines over standard schemas and standard tools
  • Coding assistants at the default abstraction level
  • Research / search / summarization over public data
  • Chatbot-shaped customer service for categories where brand intimacy is not load-bearing
  • Standard agentic workflows (email triage, calendar management, meeting prep, CRM updates)
  • Consumer "do X for me" interfaces where X is something a model can reliably do across most populations

If your business lives here, your competitive surface is distribution, trust, and audience rather than product. The tool will not differentiate you. Only the brand around it will.

The Unabsorbable Zone​

Each of these is a specific form of opinionation that a frontier lab cannot replicate without breaking its own business model:

1. Opinionated taste and aesthetic​

A lab's product is the mean. An opinionated brand is the particular. If your product's defensibility is "this looks and feels and behaves like a specific point of view," no lab can match it without diluting its own mass-market appeal. The soulful indie product is safer than it has ever been.

2. Opinionated values​

Any product whose identity is built on a clear values stance (religious, political, cultural, ethical, aesthetic) is outside the lab envelope. Labs cannot ship a product for a specific values community without alienating the others. Values-first products are wide-open territory.

3. Opinionated curation​

Deciding who belongs in a room and who does not is something labs cannot do. They cannot vouch for strangers. They cannot run hyper-curated communities. The hard judgment call of who gets in and who does not is a human act, and the product quality depends on someone with taste making that call repeatedly.

4. Specific-human bundles​

A specific named human plus an AI stack wielded on the client's behalf. Being someone's person. Taste, accountability, presence, and memory of the specific client's specific situation. A lab can ship the tools. A lab cannot be the specific human.

5. In-person, local, embodied work​

Distributed software companies cannot run an 80-person salon in a specific city on a specific night. They cannot run a church-basement workshop, a retreat in the mountains, a dinner at a chef's loft. Physical presence at a specific time with specific people is outside their operating envelope.

6. Sovereign and self-hosted infrastructure​

Anything that must run on the user's own hardware, infrastructure, or data (not the lab's cloud) is outside the lab business model. Local agents, self-hosted stacks, and open-source sovereign tools cannot be absorbed into a chat interface.

7. Human-unicorn-led enterprises​

Products built around a specific cultural leader, artist, founder, or operator whose trust, taste, and network are themselves the asset. Labs ship the infrastructure; they do not become the leader.

8. Deep custom professional services​

"Stand up the entire agentic stack for this specific business, in their industry, with their data, with their people, in a defined engagement window, and transfer ownership." That is a practitioner job, not a chat-interface job.

9. Teaching, shepherding, and mentoring​

Guiding a specific person through their specific transition, one conversation at a time, adapted to their actual context. This is a role, not a scalable product. Unabsorbable because the leverage lives in the relationship, which the transcript never fully captures.

10. Courageous or contested positions​

A frontier lab will not ship a product that gets it sued, cancelled, or regulated. It will not take contested positions. It will not build for uses whose mere existence alienates large user segments. Independent operators can. The things a lab cannot afford to build are often exactly the things a specific human with conviction can build fastest.

The Test​

Audit anything you build against a single question:

Does this product or service require a specific point of view to work?

  • If the answer is yes, it is probably unabsorbable. The labs cannot ship your opinion.
  • If the answer is no, assume a lab will ship the capability in 12-18 months and plan accordingly.

"Point of view" is the operational definition of opinionation. It includes taste, values, curation choices, a specific human's presence, a specific community's standards, a specific set of principles that exclude alternatives. Wherever a product or service depends on someone saying "this way, not that way" and standing behind the choice, labs cannot follow.

What This Means in Practice​

If you are an entrepreneur building in 2026:

  1. Audit your product against the absorbable-vs-unabsorbable test. If it lives inside the chat interface as a feature, assume a lab will ship that feature within 12-18 months.
  2. Move the value toward opinionation. Every step toward taste, values, curation, presence, or point of view is a step into territory frontier labs cannot enter.
  3. Build audience and community as durable assets. Levels's own defenses (audience and community) are correct. These are unabsorbable by definition: an audience is loyalty to a specific point of view, which labs cannot offer.
  4. Use the labs' own tools against their absorption strategy. Every primitive they ship makes it easier for a specific human with taste to build an opinionated product on top of it. Configuration-over-construction (see AI-Native by Configuration) is exactly that leverage.

If you are a practitioner building consulting, coaching, or services:

  1. Your value is going up, not down. Everything on the unabsorbable list needs humans doing the actual work. You are that human.
  2. Specialize into the opinionated zones. Industry-specific, relationship-heavy, values-aligned, in-person, sovereign-infrastructure engagements are the safest bets for the next five years.
  3. Package your taste and your track record. Those are the parts of you that cannot be absorbed.

The Meta-Frame​

Frontier labs are doing exactly what rational competitors should do. The absorption strategy is rational; it is the only strategy available to them given their incentives. No outrage required.

The correct response is clarity about where the line is. Labs will keep pulling more and more capability into their chat interfaces. Opinionated humans building opinionated products and running opinionated communities will keep thriving in the territory the labs cannot enter.

The question for every builder in 2026 is how to build something Anthropic cannot build, because they cannot afford to be you.

Further Reading​

Source​

Peter Levels (@levelsio), post on X, April 13, 2026. The absorbable-vs-unabsorbable frame and the "they won't come at it how you expect" observation are adapted from that thread with attribution.