SaaStr：放弃优先认购权、AI降本与IPO门槛提升的实战教训

The SaaStr-side lessons from Valar, Jev, Haiku 5.5, Airtable and the Decacorn boom

#1. Got cut out of a round? Say it once, then drop it.

Years ago I was the only investor in a company for years. I bridged it when no one else would and helped wherever I could. Then a round came together, I got cut to zero, and Harry got an allocation. I said something about it, because there was no good answer for why he got in and I didn’t.

It didn’t change the outcome. It just bent the relationship with the founder. Valar and Day One is the same story with Quinn Emanuel and a Delaware filing attached, and it’s already gone further than either side wanted.

My learning → If you’re the investor, you get one objection. After the founder decides, the cap table is theirs, and they have to work there for the next 20 years.

Who is right and who is wrong: Day One or Valar?

“Everyone was saying that Day One was suing their portfolio company, which was completely wrong. It is Valar that wants declaratory judgment to move on.

Sequoia came in very aggressively, and everyone had to waive their pro rata… https://t.co/7itjSRtj9p pic.twitter.com/1NBT6dCELP

— Harry Stebbings (@HarryStebbings) October 15, 2026

#2. If you ask everyone to waive, don’t carve out one investor.

Big funds put “all existing investors waive pro rata” in nearly every term sheet, and most early investors go along with it. What set Day One off at Valar was the exception: one existing investor got an allocation as a “strategic.”

If that investor is already on the cap table, you already have whatever strategic value they bring. Giving them more ownership while asking everyone else to take zero looks like playing favorites, because it is.

My learning → Founders, if you’re asking for a waiver, make it apply to every existing investor. One carve-out is what turns a routine ask into a lawsuit.

#3. Pro rata was priced into your seed round.

The most honest take I’ve heard on pro rata came from Keith Rabois a few years ago. Early investors pay up partly because they expect to keep buying in later. Take that away, and they’d pay about half the price. No founder took him up on it.

Notice also that the megafunds asking seed investors to waive never waive their own rights. That tells you what the rights are actually worth.

My learning → Founders, if you plan to strip pro rata later, remember you got a higher seed price because investors expected it. Plan for it at the seed, not at the Series B.

#4. Judge a model by what you can ship, not the benchmark.

Muse Wonder Full posted frontier-level scores, and I can’t put it in my own app. It’s only available inside Muse Code. The API version is cheaper and less capable, at roughly the same price as Opus 5.5.

That doesn’t make Meta’s work unimportant. The consumer Muse app could be a monster. But for a B2B founder deciding what to build on, a model you can’t call isn’t an option.

My learning → Ignore leaderboard results for models without an API. Test only what you can put in production this week.

#5. Re-run your model costs this month.

Haiku 5.5 is 90% cheaper than Haiku 4.5 for prompts under 100K tokens, and it’s nearly as good as Sonnet for a lot of what I run. Sonnet 5.5 is only about 10% cheaper than before, but it’s twice as fast in my testing. Speed matters more than the evals admit. Faster means you can do more in the same time.

This is the first time in two years my AI costs have actually gone down. Token prices kept falling, but usage grew faster, so none of our portfolio companies saw COGS drop. That changed this month.

My learning → Swap models and re-test every workload now. Then use the savings to run AI on every record instead of a sample.

#6. Use a decision model for your yes/no calls.

I use Jev in production. It doesn’t write text. It returns a decision and a confidence score. It’s very fast and very cheap, and it gets maybe 70% to 80% of carefully prompted decisions right today.

That covers maybe 10% of AI use cases right now: classification, routing, yes/no filtering. That’s still a lot of frontier-model spend you can move to something far cheaper.

My learning → Pull your classification and yes/no calls off frontier models. Save the expensive models for work that needs reasoning or writing.

#7. When the big company’s version is a side project, bet on the focused team.

Cloudflare, Amazon and others can build something like Jev today. It’s not that hard to match where the category is right now. The question is whether they’ll still have a real team on it in a year, once it’s a small line item next to their core business.

TypeSafe has about 20 people who do nothing else. That’s why I’m cautiously bullish at $7.5B, even with that many big competitors circling.

My learning → If a giant ships your product as a feature, ask how many people they’ll have on it in 12 months. Usually it’s not many.

#8. Selling to Bending Spoons means your team is gone. Decide with that in mind.

Evernote, Vimeo, and now Airtable. When Airtable’s deal was announced, almost everyone talked about the price and the multiple. Very few people mentioned that most of the team would lose their jobs.

Rory’s counterpoint is fair. Sometimes a clean exit is better for employees than a two-year grind where the best people leave anyway. But with Bending Spoons, you know what happens next. With Vista or Thoma Bravo you at least have a chance.

My learning → If you sell to a cost-cutter, negotiate employee economics before you sign. Afterward, you won’t have the leverage.

#9. If you’re going to cut, cut once and cut enough.

Cloudflare had a great year and was an AI beneficiary. Matthew Prince still cut 20% in one day because the company was moving too slowly. They beat the quarter and reaccelerated.

Netflix at 5% and HubSpot at 7% are different. Cuts that small don’t change how fast a company moves. They just add another round of anxiety for everyone who’s left.

My learning → One meaningful cut beats three small ones. Do it once, do it right, and get back to growing.

#10. $500M in ARR may not be enough to IPO anymore.

The B2B apocalypse ended for companies with massive scale. For everyone else it got worse, because no one wants to buy software. When we started the show, $500M in ARR was enough for a safe IPO. Today it may take $1.5B to $2B.

Rory’s math is sobering. There are a few hundred midsize software companies, and big tech can only buy about 10 a year. TechWolf shows another path. It rebuilt itself around AI and sold to SAP in Belgium’s biggest venture-backed software deal.

My learning → If you’re at $100M trying to get to $300M, assume no IPO. Go AI-first now, and build toward the strategic buyer who needs you.

Scale Is the Only Safe Place in B2B Right Now

This week had 42 decacorns in ten months, an $870M Series A 24 days after launch, and midsize software companies that can’t find buyers. All of that is true at once. Capital and attention are going to the winners and to scale. Everyone in between has to get AI-native fast, or plan to run the business for cash.

Jason’s Takes is the SaaStr AI companion to our weekly 20VC x SaaStr recap with Harry Stebbings and Rory O’Driscoll.