SaaStr 786: Adding AI to SaaS — Inside the AI Product Strategies of Figma, Cloudflare, GitHub, and Ramp

Podcast: The Official SaaStr Podcast
Air Date: January 10, 2025
Host: Dani (CEO, Jam)
Panelists:

• Mario Rodriguez (CPO, GitHub)
• Sam (VP Design, Ramp)
• Dane Knecht (SVP Emerging Technologies, Cloudflare)
• Vincent van der Meulen (Head of AI, Figma)

Overview

This episode features a high-profile panel of product and engineering leaders from GitHub, Ramp, Cloudflare, and Figma, offering an insider’s look at how major SaaS companies are integrating AI into real-world products and processes. Topics range from building “magic” experiences with AI, to adapting product roadmaps for rapid AI advancements, building multidisciplinary teams, evaluating AI for production readiness, and predictions for the future of SaaS in an AI-native world.

Key Discussion Points and Insights

1. Defining “AI” in Product — From Tech to Magic

• AI Isn’t the Product, It’s the Magic Layer: Sam (Ramp) emphasized that users don’t interact with “AI” per se—the goal is seamless, “magical” experiences where pain points evaporate (06:32).
  • Quote: “AI is not the product. AI is one of the ways in which we make people's lives easier. The main thing that I try to focus our practice around AI around is does it feel like magic?... If it feels like magic, it's AI in my book.” — Sam (06:32)
• AI as an Invisible Helper: Dane (Cloudflare) and Mario (GitHub) shared the importance of building AI that feels native, not as a bolt-on or as an intrusive chat interface.
  • Quote: “AI can become really powerful when you're building into your application not as a chat interface, but really as something that disappears and becomes part of it.” — Dane (13:46)

2. Company Deep Dives: The Biggest AI Features

GitHub — Copilot

• Success Factors: Taste, developer-obsession, and timing (right product, right moment) shaped Copilot’s trajectory.
• Turning Tech into Product: Transitioning from demo (OpenAI Codex) to value required relentless optimization (performance, UX, latency, prompt engineering).
• Vision: Lowering the barrier to software creation, extending to non-developers (“your grandmother, your child…”).
  • Quote: “If we're able to do that, I think we're going to really accelerate humanity forward.” — Mario (05:10)

Ramp — Embedded, Unobtrusive AI

• Use Cases: Automating expense reports, invoice processing, and workflow suggestions.
• Design Principles: Make the experience disappear—don’t just speed it up, remove the pain entirely.
  • Quote: “We make that technology, the experience disappear. We ask not how can we make the experience better or slightly faster, it's like, how do we make it go away?” — Sam (07:26)
• Empowering Engineers: Create low-risk playgrounds for engineers and designers to experiment rapidly with AI.

Cloudflare — AI Everywhere at the Edge

• AI Cloud: Providing primitives (databases, GPUs at the edge) for global, scalable AI app deployment.
• Infrastructure Focus: Make AI deployment invisible to developers, auto-scale, and resilient.
• AI Placement: Device, edge, or gateway-based depending on use case—GPU rollout to 170+ cities (11:56+).

Figma — AI to Complement, Not Replace Designers

• AI “Bundle”: Features like AI search, prototype creation, and automatic layer naming to remove drudgery and boost creativity.
• Iterative, User-Centered Development: Heavy prototyping, “maker weeks,” and frequent evaluation cycles.
• Quality Controls: Custom visual “eval” systems to ensure AI meets high product standards before launch.
  • Quote: “When you do evals, do your evals in a way that matches your product... For our search feature, we would have the AI search model just spit out all the search results on the Figma Infinite canvas, and then we could go in with a figma plugin to actually rate those search results as good or bad.” — Vincent (31:22)

3. Product Roadmaps in the Rapid AI Era (20:15–26:49)

• Strategic Bet-Driven Planning: GitHub and Cloudflare both use flexible, “horizons-based” roadmaps with strategic rather than tactical detail.
  • Quote: “Anything that we actually say four quarters from now is completely untrue. And if we actually are doing really good and sticking to that, then we probably played a lot of very, a significant amount of time.” — Mario (21:20)
• Constant Review and Iteration: Monthly or 4-6 week cycles to review and adjust plans.
• Hackathons as Pivot Points: Figma’s “Maker Weeks” can spawn new product directions overnight.

4. Experimentation, Evaluation, and Quality

• Tactical Experimentation: Small, empowered teams run rapid cycles to prove hypotheses and innovate (“demos” as validation).
• Offline and Online Evals: Essential both at GitHub (Copilot’s “coffee” suite) and Figma (visual eval plugins).
  • Quote: “You could experiment all day in AI and not accomplish anything without being able to understand how evals are going.” — Mario (27:57)
• Benchmarks and Product Fit: Evaluation systems must reflect real usage and quality—standard benchmarks can be gamed.

5. Speed vs. Quality: The First-to-Market Debate (32:24–36:50)

• Bias Toward Velocity: Most panelists favor “shots on goal,” frequent iteration, and learning fast over waiting for certainty.
  • Quote: “You have to put yourself out of a job every 18 months or someone else will do it for you.... Make a lot of bets very quickly.” — Sam (32:46)
• Plugins as an Accelerator: Figma leverages its plugin ecosystem to deliver rapid AI innovation, even if core features move slower.

6. Building World-Class Applied AI Teams (36:50–43:44)

• Multidisciplinary by Default: Mix of ML engineers, product engineers, designers, and researchers, often driven by intrinsic excitement about AI.
  • Quote: “A lot of these people reskill automatically because they're so passionate about AI and are just constantly building side projects in their own time.” — Vincent (39:49)
• Decentralized AI Efforts: At GitHub, “every team is a Copilot team” to avoid bottlenecks and maximize impact.
• Internal AI for Operations: Ramp and Cloudflare both highlight AI’s transformative power behind the scenes (support, ops, risk) as well as customer-facing.

7. The Next Five Years — AI-Native SaaS (45:02–47:44)

• AI-Native & Invisible: Products will evolve to become AI-native, not just “AI features” tagged on.
  • Quote: “If we really lean into that and go and extract the maximum value of natural language...then the product looks completely different.” — Mario (45:15)
• Minimal User Effort: Ramp aims for less user time—ideally, “nobody's using Ramp, it just works” (45:48).
• Blurring Job Roles: Figma predicts “role collapse”—design, engineering, product all blended (47:05).
• Back to Business Value: AI as an enabler, not an end—ideally “disappearing” from the conversation except as an integral part of value creation (Dane, 46:28).

Notable Quotes & Memorable Moments

• On Prompt Engineering and Talent
  “If you're great up on prompt engineering, I want to talk to you and hire you.” — Mario, GitHub (03:56)

• On Quality vs. Speed
  “If we start shipping AI features that don't meet your quality bar, very frustrating experience.” — Vincent, Figma (36:00)

• On Experimentation Culture
  “Giving the people in teams room to experiment is even more important in the speed AI technology is moving today...” — Dane, Cloudflare (26:15)

• On Product Vision in Five Years
  “Some level of role collapse is going to happen... designers will be able to create software, engineers will be able to create designs...” — Vincent, Figma (47:05)

Timestamps for Key Segments

| Timestamp | Topic/Quote | |---------------|--------------------------------------------------------------------------------------------------| | 02:32–06:05 | GitHub Copilot: origin, developer focus, future vision | | 06:32–11:30 | Ramp's philosophy: AI as magic, ergonomics, experimentation culture | | 11:56–14:55 | Cloudflare: AI at the edge, infrastructure for SaaS developers | | 15:01–19:13 | Figma’s AI journey: Jambots, bundles, evaluation processes, multidisciplinary teams | | 20:15–26:49 | Adapting product roadmaps for rapid AI development; Hackathons & flexibility | | 27:57–32:24 | How to evaluate and test AI features in production; offline/online evals | | 32:46–36:50 | Is it better to lead or to follow in the AI feature race? Panel on speed vs. quality | | 36:50–43:44 | Applied AI teams: composition, reskilling, product ownership | | 45:02–47:44 | Predictions: The AI-native SaaS future, role collapse, invisible AI |

Tone and Takeaways

The tone is candid, tactical, and optimistic, resonating with panelists’ direct experience. Speakers emphasize humility (learning by doing), the necessity of velocity, multidisciplinary collaboration, and a relentless focus on user value—AI as a means, not an end.

For SaaS founders and builders, the clear message:

• Stay nimble, experiment boldly, build multidisciplinary teams, and iterate evaluation fast.
• Seek to create experiences where AI “disappears”—users remember what’s magical, not what’s artificial.
• The field’s changing quickly: A year is an eternity in AI; five years will be unrecognizable.

End of Summary