AI go-to-market, pivoting, and early hiring for founders

Executive overview

Founders building AI companies face three recurring decision points: how to enter a market with incomplete AI capabilities, when to pivot away from a product that has some traction, and when to make early hires. The temptation at each point is to over-invest in structure before validating the fundamentals.

The core principle across all three: pace of learning beats pace of execution. Qualify before you segment, find conviction before you pivot, and let things break before you hire.

Hire only when things are visibly breaking; pivot only when you've lost conviction; go to market with the smallest slice that delivers real value.

AI go-to-market in legacy industries

• Three paths: sell software to the incumbent firms, start your own firm using AI, or acquire an existing firm.
• Most YC companies use path one — build a focused tool for a narrow, high-value workflow.
• Path two (starting a new firm) works if you track automation rate as the primary metric, not revenue.
• Failure mode: automate 20% of the work, then scale headcount anyway — you've just built a manual business with software on the side.
• Force the automation habit with a visible metric everyone watches; slower growth with rising automation rate signals more to investors than flat automation with growing revenue.
• Find early-adopter customers who are already incentivized to buy before the product is polished — pre-qualify hard to filter them in.

Choosing market segment: enterprise vs mid-market

• Early on, pace of learning is the primary asset — it beats potential contract size.
• Chasing enterprise from day one is like a moonshot: long cycles, slow feedback, hard to iterate.
• Exception: if the problem only exists at enterprise scale, there's no choice — but start with the smallest possible footprint inside that customer.
• Mid-market is generally the better starting segment: faster decisions, shorter feedback loops.
• Segment selection matters less than finding the right individual buyer — someone empowered to decide and incentivized to act.

When and how to use AI SDRs

• AI SDRs work when plugged into a sales process that already works.
• They don't solve a broken sales motion — using them as a last resort hasn't worked.
• The two hard problems — who to sell to, how to get their attention — must be solved by the founder first.
• Once those are solved, AI can scale the execution (outreach, follow-up, scheduling).
• Same advice as for hiring a first salesperson: only bring one in once you have a playbook.
• AI SDRs are "that advice times ten" — even less forgiving of an unvalidated process.

Investing now vs. waiting for better models

• If the current product would be made irrelevant by the next model release, it's the wrong thing to build.
• If better models make the product better, invest now — you'll learn during the build and be ready to plug in improvements immediately.
• Companies already working got dramatic step-ups when Sonnet and strong codegen models arrived; those who waited missed the compounding.

When to pivot with existing traction

• The hardest situation: some revenue, but slow growth and low conviction.
• Firecrawl (formerly Mendable) had hundreds of thousands in ARR when they pivoted — they noticed a niche internal tool they'd built for themselves was in higher demand than their core product.
• The signal isn't a formula: it's a deep conviction built from many conversations.
• Key indicator: customers don't all describe the product the same way — the relationship between what you're building and how people use it is disorganised.
• Pivoting is exhausting; confirm you have the energy to start over before committing.
• Have multiple pivot ideas in hand — evaluating one idea in isolation leads to demoralisation if it gets rejected.
• The leading indicator is often simpler: you've stopped believing the current path can get big.

Good idea vs. great idea

• There may be no such thing as a "good" startup idea — only great and not-great.
• A great idea surfaces through customers who need it every day to solve a real pain.
• If no customers describe it identically and urgently, it's not great yet.
• Test aggressively: build the wackiest version in two weeks, put it in front of people, look for signs of greatness.
• Founders with genuinely great ideas rarely call them great — they're obsessed with validation, not with the idea itself.

Pivoting away from technical difficulty

• A technically hard idea is an advantage, not a reason to pivot — high bar means fewer competitors.
• When scope feels overwhelming, reduce scope rather than abandon the idea.
• Build the simplest internal version first; use it yourself to generate early customer conversations.
• Risk: technical difficulty becomes an excuse to avoid customer contact — guard against it actively.

When to start hiring

• If you have time to think about whether to hire, it's too early.
• Right time: things are visibly breaking, schedules are overloaded, specific functions (engineering, sales, onboarding) are failing.
• Start interviewing when you see early indicators of breakage — the first hire takes three months to arrive.
• First hires almost always come from personal networks — people who already trust you and know the product.
• Hiring is not a success metric. Headcount is a cost; revenue is the metric.
• Opportunistic hires are valid — but only when there's a genuine superlative: smartest, best, most relevant background.
• Founders often hire a VP Marketing before understanding what that job requires; learn the role first.

Open sourcing enterprise SaaS

• Open source as a go-to-market is most natural for dev tools — your customers are your peers.
• For non-dev enterprise products, open source shortens sales cycles by building trust and removing data-privacy objections.
• Medplum (open source EHR) used open source not to attract developers but to compress a multi-year enterprise sales cycle.
• Self-hosting capability removes the objection "we can't send sensitive data to a startup."
• Tradeoff: self-hosting is expensive to support — price it accordingly.
• In AI products, self-hosting requests are now routine; founders who build it quickly have an edge.