Scale & Strategy
Eight Key Lessons for Rocket-Shipping AI Products
As LLM-native applications continue to evolve from early experiments into production-ready platforms, a new wave of best practices is emerging—shaped by the hard-won lessons of top AI leaders. From fast iteration to intelligent team structure, pricing discipline to go-to-market clarity, there’s a growing playbook for AI products that scale.
These eight lessons, distilled from conversations with builders at Adobe, Notion, Tome, Snorkel, Intercom, Perplexity, and OpenAI, highlight what it takes to ship AI products fast—and well.
1. Speed > Perfection
AI startups win by moving fast. With open access to foundational models and developer tools, the fastest route to product-market fit is often building on third-party models to launch quickly, test early assumptions, and iterate in real time.
Aravind Srinivas, co-founder and CEO of Perplexity, embraced this speed-first approach:
“You only have a right to think about moats when you even have something… Yes, I am a wrapper, but I would rather be a wrapper with a hundred thousand more users than having some model inside and nobody even knows who I am.”
Early momentum attracts users, investors, and talent. Even if it means being a “wrapper” product at first, proving demand comes before building defensibility.
2. Put Users in the Driver’s Seat
The most effective AI experiences position users as co-creators. Whether it’s generating content, designing assets, or organizing information, AI should be a tool that supports—not replaces—the user’s intent.
Keith Peiris, CEO of Tome, described the creation process as collaborative:
“We’re probably never going to show you one prompt, 10 pages that happens instantly… We show you an outline, you agree or edit, then a page at a time. That iterative flow matters.”
Adobe’s Alexandru Costin echoed this with a transportation analogy:
“It’s like an Uber. You don’t have to know how to drive, you just need to get to the destination… Customers guide the computer, not the other way around.”
This co-pilot model improves engagement, trust, and creative ownership.
3. Your Users Are Your RLHF
The best AI products integrate feedback loops directly into the user experience. Every click, edit, share, or prompt becomes a training signal.
At Notion, early adopters and ambassadors help surface use cases that get baked into prompt libraries. Snorkel collaborates with customers to improve labeling quality. Adobe tracks both explicit (like/dislike) and implicit signals (downloads, shares) to optimize future generations of its Firefly model.
As Keith from Tome put it:
“We should monitor time spent, feature retention… run A/B tests between models and iterate based on what users actually keep coming back to.”
Every interaction is a data point. The best teams know how to learn from them.
4. Structure Teams for Iteration, Not Just Innovation
In a talent-constrained market, team structure is a key differentiator. The best teams separate research from product delivery—allowing researchers to explore while enabling product teams to focus on UX, integration, and speed.
Adobe pairs its researchers with “tech transfer teams” to bring innovations into production. Notion splits its AI team into two parts: model development and product integration—ensuring progress on both performance and usability.
Small teams with clear handoffs move faster and stay focused.
5. Avoid the Feature Creep Trap
As usage grows, so do feature requests. But not every request deserves to be built.
OpenAI took a clear stance: prioritize what truly matters to the customer. That meant delaying dashboards and alert systems in favor of refining core product quality.
Discipline matters. A thousand small requests can easily distract from the few features that actually drive value.
6. Understand and Manage Cost Levers
AI infra isn’t cheap—and great user experiences can be expensive to deliver. Knowing when and how to manage costs is critical.
Adobe is experimenting with distillation, pruning, quantization, and even new silicon. Intercom runs A/B tests to benchmark models like GPT-4 vs GPT-3.5, routing user interactions based on performance and cost tradeoffs.
Tome manages its infrastructure to deprioritize the free tier during peak usage. Notion enforces a “fair use” policy, slowing usage if a user exceeds 30 AI requests in a 24-hour window.
Building a great experience is step one. Making it sustainable is step two.
7. Target the Right Early Users
Not every user is equal in early-stage GTM. Prioritize the segments most likely to adopt, evangelize, and expand usage into organizations.
Tome focused early on prosumers—individuals likely to use the product both personally and professionally. These users became natural champions as the company expanded into enterprise.
A clear GTM beachhead enables faster feedback and higher LTV from day one.
8. Dogfood Relentlessly
Before shipping to power users, test aggressively in-house. Internal feedback surfaces flaws, biases, and usability issues long before they hit production.
Notion pressure-tests features through internal usage before external rollout. Tome requires internal sign-off before exposing features externally.
Keith from Tome put it best:
“One of the few things a startup can export that a big company can’t is good taste. If we don’t like the output, why would our users?”
Taste, not scale, is your advantage early on. Use it.
Final Thought
AI products aren’t just software—they’re systems. The best ones are fast, flexible, user-centric, and obsessively tested. In 2025 and beyond, the winners will be those who move fast, learn faster, and never stop iterating.
Was this email forwarded to you?
That’s it for today and as always It would mean the world to us if you help us grow and share this newsletter with other operators.
Our mission is to help as many business operators as possible, and we would love for you to help us with that mission!
