The Weekly Waypoint: The AI SaaS Boom (and Bust) This Week

The Weekly Waypoint, Issue #4

This week was a rollercoaster for AI startups. Big raises, sudden shutdowns, and one lesson that every founder (and every person with an idea) should hear.

Let's break it down.

This Week's Big Story: Two Well-Funded AI Startups Just Died

What happened: Two AI startups that raised $20M+ each in 2025 announced they're shutting down this week.

PulseAI (AI meeting scheduler), shut down after 14 months. Had $24M in funding. Problem: Calendly added AI features and made them redundant overnight.
Draftly (AI writing assistant), shut down after 11 months. Had $18M in funding. Problem: ChatGPT and Claude added the same features for free.

Why it matters: These weren't bad products. They were good products that got crushed by platform companies adding the same features. This is the biggest risk in AI right now: building something the big players will eventually bundle for free.

The lesson: Don't build a product that's just a wrapper around an API. Build something with proprietary data, workflow, or audience that can't be replicated by OpenAI adding a feature.

The Good News: Three Startups Are Winning

Cursor raised $200M at $2.5B valuation, The AI code editor is now the fastest-growing developer tool ever. Why it works: it's not just "AI writing code", it understands your entire codebase and acts as a pair programmer. The moat is deep integration, not just AI generation.

ElevenLabs hit $100M ARR, The AI voice company crossed $100M annual recurring revenue this week. Why it works: they own the voice model technology, not just the application layer. Hard to replicate.

Harvey AI hit $80M ARR, The AI legal assistant is now used by 40% of AmLaw 100 firms. Why it works: deep vertical specialization. They're not "AI for everyone", they're "AI for lawyers who need to review contracts." That specificity is their moat.

What This Means If You Have an Idea

The validation playbook just got more important:

Before building anything, ask:

Could OpenAI, Google, or Anthropic add this feature in their next update? If yes, it's a feature, not a product.
Do I have a specific audience, data, or workflow that the big players can't replicate? If no, you're vulnerable.
Have I talked to 20 potential customers? If no, you're guessing.

The founders who are winning right now didn't start by building. They started by talking to customers, finding a painful problem, and validating that people would pay before writing a single line of code.

New Tools and Updates This Week

Anthropic launched "Claude for Teams", Shared workspace, shared projects, team-level permissions. If your team is paying for 5+ individual Claude Pro accounts, this is cheaper and better coordinated.

Google Sheets got AI formulas, Type "=AI" in any cell and describe what you want. "=AI(Summarize A2:A100)" actually works now. No more writing complex array formulas.

Figma AI is out of beta, Describe a design in words, get a full mockup. Designers are split: some love it for quick prototypes, others worry it devalues the craft. Both are right.

Practical: Validate Your Idea This Weekend

Got an idea for an AI product, side project, or even a career change? Try this:

Find 5 people who'd use it
Ask: "What's your biggest frustration with [topic]?"
Listen. Don't pitch.
If 3+ mention the same problem unprompted → you found something real
Show them a mockup (even just a sketch on paper)
Ask: "Would you pay $X for this?"
If 2+ say yes enthusiastically → you have validation

This costs $0 and takes a weekend. It could save you 6 months of building the wrong thing.

Not building a startup? The same framework works for any big decision. Talk to people who've done it before. Validate your assumptions. Save yourself from expensive mistakes.

The Fun Stuff

AI startup's demo video was secretly a human doing it manually, A viral "AI agent" demo turned out to be a human operator behind the scenes. The founder claimed it was "human-in-the-loop for safety" but the investors weren't amused. The term "AI washing" is now officially a thing.
Someone built an AI that roasts your LinkedIn profile, "Your summary says you're a 'results-driven synergistic leader.' The AI says you're a middle manager who uses too many buzzwords." It's mean, it's funny, and it's weirdly accurate.
The "ChatGPT diet", A nutritionist went viral for prompting ChatGPT with "I have chicken, rice, and broccoli. I'm tired. Make me 5 dinners that take under 20 minutes." The recipes were actually good. The meal planning industry is quietly panicking.

What I'm Watching Next Week

Apple's WWDC rumors are heating up. Word is they'll announce deeper AI integration across macOS, not just summarization, but agents that can control apps and automate workflows on your Mac. If true, this brings AI automation to millions of people who've never heard of Make.com or Zapier.

Also: Anthropic is reportedly launching "Claude for Education", a specialized version for teachers and students. Could reshape how AI is used in schools.

Deep Dive This Week

Pro members go deeper: "Building an AI SaaS: From Idea to Revenue", the complete guide to validating, building, and launching an AI-powered SaaS product. We covered the cautionary tales this week; the deep dive covers the exact validation framework, technical architecture, and go-to-market strategy that separates the startups that survive from the ones that don't.

Upgrade to Pro →

The best AI product idea is the one you've validated before you've built. The worst is the one you've built before you've validated.

- James

P.S. - Have an AI idea you've been sitting on? Hit reply and describe it in 2 sentences. I'll tell you if I'd validate it or skip it.