The Disposable Product Manifesto

• 11 min read

tl;dr: AI is eating product categories faster than anyone can build moats around them. Ship disposable, ship fast, run bets in parallel. The safe harbors left are regulated (finance, healthcare, defense) or human (distribution, taste, trust). VC math doesn’t math anymore. Nobody knows what’s next, me included.

Build Like It’s Already Obsolete

Most digital products one can ship from now on are structurally ephemeral. The tech they’re built on iterates in daily cycles, not monthly or yearly ones anymore. Something drops on a Tuesday and by Friday half your roadmap is a joke.

Hard truth: whatever you build probably has only months of shelf life before it goes obsolete or gets eaten by one of the big boys, such as Anthropic, which has been running this Pac-Man strategy harder than anyone else (see ANTHROPIC & THE PAC MAN STRATEGY).

Sounds catastrophic? It’s just another tectonic shift. Amazon ate small businesses across the world in 20 years. These guys will do it in 2.

Dead Man Walking

The same Pac-Man dynamic playing out at Anthropic is about to repeat across every layer of the stack. Some categories are already toast, others have a few quarters left, but the pattern is the same: anything that can be reframed as a feature inside a foundation model’s product surface, will be.

  • Wrapper startups die first: single-prompt tools, “ChatGPT for X,” thin API resellers. Zero defensibility, first to go.
  • Vertical SaaS gets gutted: anything that’s “spreadsheet + workflow + reports” becomes a Claude/ChatGPT skill you summon in 10 seconds.
  • Dev tools consolidate: IDEs, code review, testing, deployment, all folding into Claude Code, Cursor, Copilot.
  • No-code tools collapse: Figma, Webflow, Lovable, v0 squeezed from above by native generative UI.
  • Headcount inversion: solo operators with AI leverage outproduce 20-person startups. Series A rounds for “AI productivity tools” start looking absurd.
  • VC thesis breaks: 7-year hold + 10x return math doesn’t survive a 6-month product half-life.
  • Moats shift from tech to distribution/data/trust: the only defensible plays own a proprietary data loop, a regulated relationship, or a chosen brand.

Each of the above has already started playing out. See DEAD MAN WALKING.

Now What?

SHIP AT LUDICROUS SPEED

“Move fast and break things” just hit a whole new gear. Overthinking is the new technical debt. The cost of a wrong decision has collapsed because the half-life of any decision is six months anyway.

STAY LOOSE ENOUGH TO PIVOT OR PULL THE PLUG

Treat every product like a summer fling. The moment a core provider drops the same thing natively (and they will), pivot the angle, dig deeper into a niche they won’t bother with, or kill it entirely. Act like a momentum trader, not a buy&holder.

PARALLALIZE LIKE CRAZY

What used to take a team of 5 and six months can now be built by 1 person in a week to a month. Don’t bet the house on one product, run 5, 10, 20 bets in parallel. Treat your portfolio like a VC, not a craftsman. Most will die. All might. That’s fine, you’re trading the spike, not the long hold.

REGULATED HARBORS

Not everything moves at AI speed. Some sectors are armored by stuff foundation models can’t touch: regulators, licenses, capital requirements, political patronage, and the kind of trust that takes decades to earn and seconds to lose. Pac-Man can chew through SaaS all day, but he chokes on red tape and lobbyists. If you want to build something with a shelf life longer than two quarters, this may be where you should go hunting.

  • Regulated finance: banks, payment rails, stablecoin infra, clearinghouses. Licensing, compliance, and nation-state buy-in form a wall outsiders don’t climb.
  • Healthcare: EHRs, clinical systems, anything touching HIPAA or FDA. Liability and certification kill fast iteration cold.
  • Defense & gov contracting: clearances, FedRAMP, multi-year procurement cycles. The barrier to entry is the entire point.
  • Legal infrastructure: court systems, e-filing, bar-regulated practice software. Jurisdictional moats by design.
  • Energy & utilities: grid software, SCADA, regulated monopolies. Critical infrastructure protection means no one ships fast here, ever.
  • Identity & KYC: government-tied verification, eIDAS, passport and biometric systems. State-sanctioned or you don’t exist.
  • Heavy industrial / OT: factory control, aviation, rail, maritime. Safety certifications take years and cost millions.
  • Insurance: state-by-state regulation, actuarial licensing, capital requirements. Fifty regulators, fifty headaches.
  • Tax & accounting infrastructure: direct integrations with tax authorities, audit-grade systems. Regulator-blessed or bust.

WHO’S MOONING

Not everyone’s getting eaten, though. Some roles are quietly mooning while the AI panic plays out. Distribution is king: anyone who owns an audience, a community, or a sales pipeline is suddenly the scarcest resource in the stack, because building products is cheap and reaching people isn’t. For example, think creators, growth operators, B2B founders with warm networks, and niche newsletter owners.

Same goes for taste-makers and brand builders, since differentiation can’t be prompted. And anywhere trust is the product (lawyers, accountants, doctors, fiduciaries), the human stays in the loop because the liability does too. AI made the build trivial. It made everything around the build more valuable.

That said, the Pac-Man is circling here too. clay.com is gunning for growth operators, heygen.com lets anyone fake the creator look without the creator, beehiiv.com is baking AI writing into the newsletter stack, harvey.ai is coming for lawyers, hippocraticai.com for doctors. None has killed the role yet, but they’re compressing the floor. The moat isn’t the function anymore; it’s the relationship, the reputation, and the judgment behind it.

More receipts in WHO’S MOONING.

A16Z, WE NEED TO TALK

I mentioned earlier that the VC thesis will break, and I’d like to extend on this because it’s been intriguing me quite some lately. The 7-year hold, 10x exit playbook was built for a world where moats compounded. In an AI-SaaS age with 6-month product half-lives, the math just doesn’t math. And the funding itself becomes a liability: VC money demands you hire fast, scale fast, defend a moat, exactly the opposite of what survival now rewards.

So what replaces it? Honestly, no idea. Maybe bootstrapped portfolios where the founder is their own VC. Maybe revenue-based financing that doesn’t need exits. Maybe “indie VC” with smaller checks, shorter horizons. Or maybe the asset class shrinks back to deep tech, biotech, defense, and quits the app layer entirely. None feels obviously right yet. But the current model? Running on fumes.

DON’T QUOTE ME

A lot of what I just wrote probably won’t age well. The whole point is that nobody can see past the next quarter or two anymore, me included. By the time you read this, half of it is probably already wrong, and that’s kind of the thesis. Build accordingly. Or don’t. We’ll all find out together.

Notes & References

ANTHROPIC & THE PAC MAN STRATEGY

  • Claude for Chrome → ate the value prop of Browser Use, Browserbase’s agent layer, and put pressure on Perplexity Comet. (Research preview Aug 2025, GA all paid plans Dec 2025.) Source.
  • Claude for Excel → absorbed what Shortcut (formerly GoCharlie), Numerous.ai, and Rows AI were building entire companies around. (Beta Oct 2025, Pro rollout Jan 2026.) Source.
  • Claude Cowork → swallowed the desktop-agent niche Manus and OpenAI’s ChatGPT Agent were racing to own. Source.
  • Claude Design (Apr 2026) → directly torched the prompt-to-UI category occupied by Lovable, Vercel v0, and Bolt.new. Figma stock dropped ~7% the day it launched, Mike Krieger had quit Figma’s board 3 days prior.
  • Claude Skills → reframed the entire “Custom GPT” / agent-marketplace concept that OpenAI’s GPT Store was built on, and made plenty of single-purpose wrapper startups redundant overnight.
  • Native memory in Claude Code & Cowork (“Auto Memory” + “Auto Dream”) → eats into Mem0’s entire reason for existing as a third-party memory layer.
  • Imagine with Claude / generative UI in chat → quietly absorbed what Vercel’s generative UI SDK and a wave of “AI-native interface” startups were positioning around.

DEAD MAN WALKING

  • Wrapper startups die first → Jasper AI: built on GPT-3, hit unicorn status, then ChatGPT launched and cut their valuation by ~$300M, forced a 30%+ ARR forecast cut, layoffs, and a CEO replacement in under a year. Source.
  • Vertical SaaS gets gutted → The workflow layer above CRM/ERP is detaching from the UI: AI agents now talk to CRM, data warehouse, and email simultaneously without clicking anything, eroding the surface vertical SaaS was built on.
  • Dev tools consolidate → Claude Code hit $2.5B ARR within ~8 months of launch; Cursor hit $2B ARR by Feb 2026 and is in talks with SpaceX for a $60B acquisition option. Source.
  • No-code tools collapse → Figma stock fell 6.8% the day Anthropic launched Claude Design (April 17, 2026) and is down 80%+ from its post-IPO high.
  • Headcount inversion → 73,000+ tech employees laid off in 2026 YTD. Block, Snap, and Atlassian all explicitly cited AI efficiency. Source.
  • VC thesis breaks → a16z declared in Jan 2025 that the 20-year SaaS golden rule (per-seat pricing for streamlining human tasks) is “no longer valid.”
  • Moats shift from tech to distribution/data/trust → a16z’s portfolio shows companies with proprietary-data moats are setting the fastest records from $0 to $100M ARR.

REGULATED HARBORS

  • Regulated finance → Neobanks N26, Revolut, and Trade Republic kept hitting regulatory fines as they scaled, while incumbent banks compounded. The licensing layer is the moat.
  • Healthcare → Epic now controls 55% of US acute care hospital beds and won ~70% of new hospital contracts in 2024. Three years after Oracle bought Cerner for $28B trying to disrupt it, Oracle Health lost 57 acute care customers.
  • Defense & gov contracting → Palantir ($30.5B valuation) had to literally sue the federal government to get evaluated as vendors. That’s the entry barrier, and it’s exactly what protects them now.
  • Energy & utilities → Banks alone need months of model risk reviews before deploying any AI tool, even low-risk ones (per the Bank Policy Institute). Utilities are slower still. AI’s “ship Tuesday, iterate Friday” cadence is structurally incompatible with critical infra.
  • Insurance → Lemonade has not been profitable since founding, despite being the poster child for AI-native insurance. Hippo only turned its first quarterly profit in Q4 2024 after a decade. Capital requirements + state-by-state licensing = AI doesn’t help you skip the hard part.

WHO’S MOONING

Growth ops & outbound sales

  • clay.com: data enrichment + AI workflows for outbound prospecting at scale
  • apollo.io: B2B contact database + AI-powered email sequencing
  • artisan.co: “AI BDR” (Ava) that runs full outbound cycles autonomously
  • outreach.io: sales engagement + AI agent layer for the full deal cycle
  • 11x.ai: AI sales reps positioned as headcount replacement
  • amplemarket.com: all-in-one AI sales platform (prospecting, outreach, enrichment)
  • aisdr.com: autonomous AI SDR running personalized outreach end-to-end
  • coldiq.com: cold outbound systems and AI-driven lead-gen infrastructure

Enterprise sales leadership & forecasting

  • gong.io: conversation intelligence, deal coaching, win-rate prediction
  • clari.com: pipeline forecasting and revenue execution
  • zoominfo.com: account intelligence + intent signals via Copilot

Workflow automation & AI agents

  • relay.app: human-in-the-loop AI workflow automation across SaaS tools
  • lindy.ai: no-code AI agents for email, meetings, and ops workflows

Creators & video

  • heygen.com: AI avatars and video generation from text
  • captions.ai: AI editing, dubbing, and caption-driven video production
  • opus.pro: AI clipping of long-form video into shorts
  • descript.com: edit video and audio by editing the transcript

Newsletter operators & writers

  • beehiiv.com: newsletter platform with AI writing, growth, and ad tools baked in
  • substack.com: adding AI features around recommendation and growth loops
  • lex.page: AI-native writing surface for long-form

Lawyers

Accountants & bookkeepers

  • pilot.com: outsourced bookkeeping + AI back office for startups
  • digits.com: AI-native accounting and finance dashboards
  • puzzle.io: real-time accounting with AI-driven categorization

Doctors & clinicians

Payments & commerce