BusinessNotes/wiki/concepts/ai-market-shift.md

# AI Market Shift

#concept #ai #positioning

## Summary

AI is simultaneously the thing that destroys the hourly service model, the tool that removes the cost of building offers, and — least obviously — an instrument for qualifying buyers. The sources treat AI as a market restructuring rather than a product category. This page is the **market-side** view; its labor-side twin — how AI restructures teams, roles, and careers — is [[future-of-engineering-work]].

## Current Understanding

**Three distinct roles, worth keeping separate:**

| Role | Claim | Source |
|---|---|---|
| **Substitute** | The buyer has a ~$200/mo AI alternative → competing on rate zeroes margin; hourly development is dead | [[2026-06-15-rodenko-selling-development-expensively]] |
| **Complement** | For every $1 spent on AI, a company spends ~$6 on specialists to configure it — AI opens a new funnel rather than taking work | [[2026-06-15-rodenko-selling-development-expensively]] |
| **Tool** | AI does the deck-and-doc drudgery; the leverage that used to require a paid coach is now free | [[2026-07-17-design-the-perfect-offer]] |
| **Filter** | Give a working AI skill away free 30 days; the reaction qualifies the buyer (below) | [[2026-06-15-rodenko-selling-development-expensively]] |
| **Leveler** | Domain knowledge on demand — "whatever business you want to be in, Claude will also tell you how to do it"; collapses the no-expertise barrier to *entering* a market | [[2026-07-29-start-a-business-with-claude-code]] |

The substitute and complement claims sit together coherently: AI eats undifferentiated execution and creates demand for the judgment that directs it. That is the whole strategic argument of this vault in one line — and it is why [[niche-selection]] says go where AI is weak and [[productized-service]] says stop selling execution.

**AI as a qualification filter** — origin [[2026-06-15-rodenko-selling-development-expensively]] (a founder gives a free AI-skill on call one; of 10 calls, 5 return, 1 closes a $100k project). Don't hide the AI; weaponize it. Give a working AI skill away free for 30 days and read the reaction:

| Reaction | Meaning |
|---|---|
| "It works" | Your value isn't needed. **Not a client.** |
| "We never got to it" | The task has no budget. **Not a client.** |
| "Weak, needs specifics" | **Target client.** |

This inverts the usual anxiety. The fear is that AI reveals your work to be replaceable; the filter accepts that for some buyers it *is*, and uses the giveaway to find the buyers for whom it isn't — cheaply, before you've spent a sales cycle on them. It also does [[pain-discovery]]'s job: the "needs specifics" answer *is* the specification.

**The demand-side sequel: if products are cheap, attention and distribution are the scarcities** ([[2026-07-26-how-to-build-a-billion-dollar-company-2027]], [[oskar-hartmann]], added 2026-07-26). In the vibe-coding era "everyone has 10–15 great products sitting in Cloud Code," so the differentiator moves from building to **distribution** ([[sales-channel-as-moat]]) — the same execution-is-commodity premise the RU sources state for services, restated for products by a third tradition. AI also *amplifies the noise it created*: 2,000 unread LinkedIn messages ("agents mail for everyone"), every channel saturated, attention "the most expensive resource on the planet" — which quietly strengthens [[relationships-as-moat]]'s case that AI-floodable channels devalue first (he independently notes offline events becoming *more* economical). And on the capital side, the same shift concentrates money: Anthropic/OpenAI's ~$4B private-equity joint ventures are a classic **land grab** (install the technology into hundreds of thousands of portfolio companies through every channel at once), while the top labs "suck nearly all free cash" out of the venture market — for everyone else, venture effectively closes ([[venture-fit]]).

**The leveler role, and an odd alliance with the adversarial evidence** (added 2026-07-29). The claim ([[2026-07-29-start-a-business-with-claude-code]], a 42s [[dan-martell]] short — attributed same day; slogan-grade, and his first claim in this page's table) is that AI removes "I don't have experience in that industry" as a barrier to starting. Note the geometry against [[ai-productivity-evidence]]: the measured gains are **novice-tilted** — which cuts *against* the expert-productivity hype but mildly *for* this role, since a founder entering a new domain is exactly a novice in it. The same evidence still contests the short's step 5 ("then ask Claude to build the product"). Unresolved whether the leveler holds outside commodity digital services (regulated/physical domains).

**What's scarce now.** Not the ability to produce the artifact, but knowing which questions to feed the tool ([[2026-07-17-design-the-perfect-offer]]). Corroborating detail: the AI's offer recommendation (workflow + dashboard + agent) independently matched the human coach's best-practice structure — the tool is a legitimate co-designer, not a novelty.

## Evidence

- **Primary:** ~$200/mo substitute; $1:$6 AI-to-specialist spend ("AI sells servicing, not product"); the free-30-day AI-skill filter with the $100k close; "narrow niche = where AI is powerless" — [[2026-06-15-rodenko-selling-development-expensively]]
- The $1:$6 ratio also appears in [[2026-06-15-making-money-with-ai-2026]] and via the meta-analysis; solo-founder revenue figures (Rezi $293k/mo) there feed the "team size no longer signals seriousness" thread
- Labor-side of the same shift: coding cost → 0, "the game is how you use it" — [[2026-07-06-sebastian-interview-ai-and-software-engineering]]
- **Counter-evidence (adversarial):** AI's measured effect is modest and often negative for experts, trust is low/falling — the substitute premise is weaker than claimed — [[ai-productivity-evidence]] / [[2026-07-18-ai-productivity-adversarial-evidence]]
- Condensed restatement — [[2026-06-15-selling-development-services-in-the-ai-era]]
- Vibe-coding product glut ("10–15 products in Cloud Code"); AI noise (2,000 LinkedIn messages); attention as scarcest resource; the ~$4B PE land grab; capital suction — [[2026-07-26-how-to-build-a-billion-dollar-company-2027]] ([[oskar-hartmann]]) — note the product-glut claim shares the commoditization premise [[ai-productivity-evidence]] contests
- "AI does the deck-and-doc drudgery — the leverage that used to require hiring a coach is now free. What's scarce is knowing which questions to feed it." — [[2026-07-17-design-the-perfect-offer]]
- "The AI-generated recommendation matched the actual best-practice structure — the tool is a legitimate co-designer." — [[2026-07-17-design-the-perfect-offer]]
- Leveler role ("Claude will also tell you how to do it"; AI as full GTM stack) — [[2026-07-29-start-a-business-with-claude-code]] ([[dan-martell]], no evidence offered; see [[claude-code]])

## Related Pages

- [[future-of-engineering-work]] — the labor-side twin of this page
- [[niche-selection]] — AI weakness defines where the defensible niche is
- [[productized-service]] — the substitution pressure is the argument for productizing
- [[pain-discovery]] — the free-giveaway filter doubles as discovery
- [[pricing-from-value]] — a $200/mo substitute is the floor rate competition collapses toward
- [[ai-productivity-evidence]] — the empirical test of whether AI is the substitute this page assumes
- [[claude-code]] — the named tool behind the leveler and build-engine roles
- [[overview]]

## Contradictions / Uncertainty

- **The $1:$6 ratio is now attributable but still uncited at root.** It appears in Rodenko and the make-money-AI video, but neither gives a primary source — it reads like an analyst statistic repeated between creators. `Status: tentative` — do not quote as fact.
- **The ~$200/mo substitute is asserted, not demonstrated.** Whether it genuinely substitutes for a dev contract or only for its commodity floor is exactly the question that decides how much of this vault's advice applies — and no source tests it. [[2026-07-06-sebastian-interview-ai-and-software-engineering]] complicates it: at the enterprise scale the substitute isn't even installable (locked-down VMs), so "the buyer has a $200/mo alternative" may hold for SMB and not for regulated enterprise.
  - **Update (2026-07-18) — empirical evidence now tests this, and it tilts toward *complement*, not *substitute*.** [[ai-productivity-evidence]] (via the vault's first adversarial source, [[2026-07-18-ai-productivity-adversarial-evidence]]) finds AI's measured productivity effect is modest (≈14–19% average) and, for experienced developers on familiar code, *negative* (METR RCT: −19%), with only ~3% of developers highly trusting AI output and ~46% distrusting it. A tool that slows experts and that most developers distrust is not obviously a drop-in replacement for a developer — it substantiates the doubt this bullet already raised, and lends weight to the **complement** row of the table above ($1:$6) over the **substitute** row. Caveats: the evidence is early-2025-scoped and novice-tilted, so it bites hardest on the *high* end (experts) and says least about the commodity floor the substitute claim targets.
- **"Go where AI is powerless" assumes a static frontier.** The sources are dated mid-2026 and mostly ignore that today's AI-weak niche may not be next year's — the one concrete durable example (COBOL, no training data) is in the Sebastian interview, not the sales sources.
- The 30-day giveaway filter has an unexamined cost: giving the skill away to the "it works" segment means handing your product to non-clients for free. Whether that's a real loss depends on whether they'd ever have paid — plausibly not, which is the point.

## Next Questions

- What's the source of the $1:$6 figure, and does it hold outside enterprise?
- How fast is the AI-weak frontier actually moving in the user's target niche?
- What is the smallest useful "AI skill" that can be given away in the 30-day filter without giving away the method?