AI Changed SaaS (And Nobody Even Realizes)

Source: Adam Robinson | https://www.youtube.com/watch?v=kS9ii5vYbOM Duration: 4 min | Published: 2026-04-09 Processed: 2026-04-10

Core Concepts

The viral anecdote that opens the video: someone gave Claude Code the domain typeform.com and four hours later Claude Code had rebuilt Typeform feature-for-feature. Used as evidence that reproducing existing SaaS UIs is now trivial.
The common counter-argument (from Dharmesh Shah, HubSpot CTO): enterprises will never trust a revenue engine to something vibe-coded in a weekend. Robinson concedes this is correct but says it misses the real threat.
- Vibe-coding = the informal term for building software by describing what you want in natural language and letting an AI coding agent (Claude Code, Cursor, Lovable) generate it.
The real structural change: the engineering-talent filter that used to gate software building has collapsed.
- Previously you needed capital, engineering hires, and time to compete with a SaaS incumbent. That triad was the natural ceiling on competition.
- Now almost any code can be written by almost anyone anywhere. The cost of entry has collapsed to near zero.
Two consequences of the collapsed filter:
- Fact 1: It is a far worse environment for selling software than before, because there is no longer a moat around the act of building the product itself.
- Fact 2: The pool of potential founders who can now credibly attempt to build a SaaS competitor is ~10,000x larger. Most will fail, but a small fraction will be exceptional (Elon-tier) and the aggregate threat to incumbents is real.
Why SaaS stocks are actually getting hurt, per Robinson: not because Lovable will replace HubSpot, but because every public SaaS now faces a vastly larger top of the funnel of potential competitors.
Three specific problems for the future of SaaS:
- Problem 1: Everyone will build everything. When a year-old SaaS product can be rebuilt in an afternoon, the product itself is no longer the moat.
- Problem 2: Every SaaS company will pivot to "agents that do the job" rather than "software that helps you do the job". Selling outcomes, not tools. Reference case: Intercom's Fin (an AI support agent that closes tickets rather than helping humans close them) — Eoghan McCabe has been public about the pivot.
- Problem 3: Agent businesses will not carry SaaS gross margins.
  - Traditional SaaS: once the software is built, each new customer is near-pure profit — marginal cost per customer ~0.
  - Agents: real per-task token costs (the LLM inference bill) and real per-interaction overhead. Revenue can look fine while gross margin structurally deteriorates month over month as usage grows.
The only defensible moats left, per Robinson:
- A data advantage (proprietary first-party data nobody else has).
- A network effect (value compounds with each user).
- Something that took years to build that cannot be vibe-coded over a weekend (deep integrations, regulatory work, distribution).
Robinson's own play (RB2B / RVDB): push the LinkedIn profiles of anonymous website visitors directly into Slack. The moat is the first-party identity-resolution dataset, not the UI. Five-minute setup.

Buildable Ideas

Data-first positioning for any new AI product — the core question becomes "what data do I have or can I collect that nobody else can?" not "what feature can I build?". Relevant to every Polerie offer.
Outcome-priced agents as a revenue model — instead of seats or monthly subscriptions, charge per closed ticket, per qualified lead, per published article. Matches the Fin and Intercom pattern and survives the margin squeeze better if pricing is set above true variable cost.
Margin-aware agent pricing model — build any agent product with a per-interaction cost line explicitly surfaced in the pricing logic. Avoids the "paper revenue, dying margins" trap.
Reproduce-a-SaaS-in-a-day exercise as a diagnostic — take a competitor's public product, rebuild it in a weekend with Claude Code, use the exercise to prove the UI layer has no defensibility and force the real moat question.
First-party data flywheels — any workflow that touches the customer (forms, onboarding, email, chat) is a data-collection surface. Treat every touchpoint as an opportunity to widen the data moat.
Vibe-coded validation layer — the speed of reproduction means market research is now faster than traditional research. Build the competitor in a day, test with real users, decide whether the category is defensible before investing further.

Key Takeaways

The threat to SaaS is not that weekend vibe-coders will replace HubSpot. The threat is that the number of credible competitors for every SaaS category just went up by four orders of magnitude.
The product is no longer the moat. Proprietary data, network effects, and genuinely hard-to-replicate assets are the only remaining defences.
The industry pivot is from "tool that helps you do the job" to "agent that does the job and charges for the outcome" — the Intercom Fin model.
Agent margins are structurally worse than SaaS margins. Any new agent business must be priced with per-interaction cost built into the model from day one.
For Jo: the Polerie thesis (own the data, own the agents, own the infrastructure) is the correct response to this shift. Every client engagement should be framed as "what data moat are we building while we solve your immediate problem?".

# AI Changed SaaS (And Nobody Even Realizes) **Source:** Adam Robinson | https://www.youtube.com/watch?v=kS9ii5vYbOM **Duration:** 4 min | **Published:** 2026-04-09 **Processed:** 2026-04-10 --- ## Core Concepts - The viral anecdote that opens the video: someone gave Claude Code the domain typeform.com and four hours later Claude Code had rebuilt Typeform feature-for-feature. Used as evidence that reproducing existing SaaS UIs is now trivial. - The common counter-argument (from Dharmesh Shah, HubSpot CTO): enterprises will never trust a revenue engine to something vibe-coded in a weekend. Robinson concedes this is correct but says it misses the real threat. - Vibe-coding = the informal term for building software by describing what you want in natural language and letting an AI coding agent (Claude Code, Cursor, Lovable) generate it. - The real structural change: the engineering-talent filter that used to gate software building has collapsed. - Previously you needed capital, engineering hires, and time to compete with a SaaS incumbent. That triad was the natural ceiling on competition. - Now almost any code can be written by almost anyone anywhere. The cost of entry has collapsed to near zero. - Two consequences of the collapsed filter: - Fact 1: It is a far worse environment for selling software than before, because there is no longer a moat around the act of building the product itself. - Fact 2: The pool of potential founders who can now credibly attempt to build a SaaS competitor is ~10,000x larger. Most will fail, but a small fraction will be exceptional (Elon-tier) and the aggregate threat to incumbents is real. - Why SaaS stocks are actually getting hurt, per Robinson: not because Lovable will replace HubSpot, but because every public SaaS now faces a vastly larger top of the funnel of potential competitors. - Three specific problems for the future of SaaS: - Problem 1: Everyone will build everything. When a year-old SaaS product can be rebuilt in an afternoon, the product itself is no longer the moat. - Problem 2: Every SaaS company will pivot to "agents that do the job" rather than "software that helps you do the job". Selling outcomes, not tools. Reference case: Intercom's Fin (an AI support agent that closes tickets rather than helping humans close them) — Eoghan McCabe has been public about the pivot. - Problem 3: Agent businesses will not carry SaaS gross margins. - Traditional SaaS: once the software is built, each new customer is near-pure profit — marginal cost per customer ~0. - Agents: real per-task token costs (the LLM inference bill) and real per-interaction overhead. Revenue can look fine while gross margin structurally deteriorates month over month as usage grows. - The only defensible moats left, per Robinson: - A data advantage (proprietary first-party data nobody else has). - A network effect (value compounds with each user). - Something that took years to build that cannot be vibe-coded over a weekend (deep integrations, regulatory work, distribution). - Robinson's own play (RB2B / RVDB): push the LinkedIn profiles of anonymous website visitors directly into Slack. The moat is the first-party identity-resolution dataset, not the UI. Five-minute setup. ## Buildable Ideas - Data-first positioning for any new AI product — the core question becomes "what data do I have or can I collect that nobody else can?" not "what feature can I build?". Relevant to every Polerie offer. - Outcome-priced agents as a revenue model — instead of seats or monthly subscriptions, charge per closed ticket, per qualified lead, per published article. Matches the Fin and Intercom pattern and survives the margin squeeze better if pricing is set above true variable cost. - Margin-aware agent pricing model — build any agent product with a per-interaction cost line explicitly surfaced in the pricing logic. Avoids the "paper revenue, dying margins" trap. - Reproduce-a-SaaS-in-a-day exercise as a diagnostic — take a competitor's public product, rebuild it in a weekend with Claude Code, use the exercise to prove the UI layer has no defensibility and force the real moat question. - First-party data flywheels — any workflow that touches the customer (forms, onboarding, email, chat) is a data-collection surface. Treat every touchpoint as an opportunity to widen the data moat. - Vibe-coded validation layer — the speed of reproduction means market research is now faster than traditional research. Build the competitor in a day, test with real users, decide whether the category is defensible before investing further. ## Key Takeaways - The threat to SaaS is not that weekend vibe-coders will replace HubSpot. The threat is that the number of credible competitors for every SaaS category just went up by four orders of magnitude. - The product is no longer the moat. Proprietary data, network effects, and genuinely hard-to-replicate assets are the only remaining defences. - The industry pivot is from "tool that helps you do the job" to "agent that does the job and charges for the outcome" — the Intercom Fin model. - Agent margins are structurally worse than SaaS margins. Any new agent business must be priced with per-interaction cost built into the model from day one. - For Jo: the Polerie thesis (own the data, own the agents, own the infrastructure) is the correct response to this shift. Every client engagement should be framed as "what data moat are we building while we solve your immediate problem?".