How to Unfuck Your SaaS by Einar Vollset

Rob's talk covered defense. Einar's covers offense. Disruption hurts incumbents, but it's also the opening that smaller, bootstrapped companies need. In stable markets, the companies with the most resources win. In disrupted markets, upstarts get their shot. Einar argues the biggest opportunity for most SaaS founders is to stop selling tools and start selling outcomes, by moving into labor budgets.

Three offensive moves

1. Move faster. This is table stakes. Mature SaaS companies are shipping features about 3x faster than they did before AI. Use that extra speed to build toward the moats Rob described, not just to add more features.
2. Charge the agents. For some products, AI agents will become users. Figure out how to give agents easy access and how to charge for it. Discovery is splitting too: humans and agents may find products in different ways.
3. Move into labor markets. This is the big one. Businesses spend far more on salaries than on software. A lot of SaaS works like this: the owner buys a tool, and an employee clicks around in it to get a result. Someone is going to sell your customers that result directly, without anyone doing the clicking. Einar's point is that it should be you. You already have the customer relationships, industry knowledge and trust. A well-funded startup with a huge AI budget and no customers is in a worse position than you.

Why building features alone won't protect you

The classic product loop still matters: spot a customer need, ship it, get feedback, repeat. AI only speeds up the coding part, and everyone gets the same speedup. You might spend months designing the perfect feature, and a competitor can study your product and copy it in an afternoon. Shipping features is necessary, but it isn't a moat.

The real moat: a learning loop

Moving into labor markets alone isn't defensible either. If you build an AI receptionist, a flood of competitors will build one the next day. What makes it defensible is a product that improves itself from real-world outcomes:

• The product takes an action, like replying to a ticket or chasing an invoice.
• It observes the result: did the customer pay, were they happy?
• That result improves the next action, ideally automatically.

The best learning loops:

• Stay with you. The learning doesn't leak out, and customers can't export the value.
• Require real-world time and experience. Competitors can't simulate their way to catching up.
• Improve results across customers. A competitor would have to live through years of your customers' experience to match you.

That's what keeps you ahead as AI models improve. Nobody can plug next year's model into a fresh product and instantly match what yours has learned.

Example: how a collections SaaS could evolve

1. Dashboard. Tracks who owes what and which offers were made.
2. AI assistant. Drafts reminder emails for a human to review and send.
3. AI worker. Decides on its own who to contact and what offer to make. This is where it starts competing for labor budgets.
4. Compounding AI worker. Does all of that and gets better with every outcome.

Five tests for your product

• Does your product take the action, or just recommend it?
• Can it see what happened afterward?
• Does the outcome change what it does next?
• Can what it learns improve results for other customers?
• Does the accumulated learning stay with you?

Takeaway: Ship faster, but point that speed at real moats. Look for the work your customers currently pay people to do in your product, and build toward doing that work yourself. Then make sure your product learns from every outcome, so your lead grows instead of shrinking as AI gets better.