By playing, you agree to uListen's Terms and Privacy Policy and YouTube's Terms of Service.

First Round Capital

How Gamma pulled off their AI pivot | Jon Noronha (Co-founder and CPO of Gamma)

In this episode of In Depth, Brett sits down with Jon Noronha, co-founder and CPO of Gamma, the AI-native presentation platform used by more than 100 million people.

Jon NoronhaguestBrett Bersonhost

Jul 24, 20261h 4mWatch on YouTube ↗

Summaries, chapters, transcripts, and FAQs on this page are made by uListen, not YouTube.

CHAPTERS

Jump to publisher timestamps

Chapter summaries

Gamma’s AI pivot: solving the blank page and chasing sudden demand
Jon Noronha and host Brett Berson open with the core story: Gamma’s big inflection came when it focused on the blank page problem and bet heavily on generative AI. Jon describes the feeling of true product-market fit as shifting from pushing growth uphill to chasing it downhill as demand outpaced the team’s capacity.
- •Jon says people were begging to pay even before Gamma monetized
- •Jon describes the AI bet as a company-wide “Hail Mary” that paid off
- •Brett frames Gamma’s bet as focusing the entire company on the blank page problem
- •Jon shares the “chasing the rock downhill” definition of product-market fit and the operational chaos that came with it
Three phases of (not) finding fit: prototypes, weak traction, then an AI-driven breakout
Jon lays out Gamma’s path from having no product-market fit, to middling early traction, to a major inflection after the AI launch. He explains how the company was born in the pandemic era with a remote-work thesis, then later found a stronger “why now” with AI.
- •Jon calls 2020 the “none era” of product-market fit, driven by remote work as the initial thesis
- •Gamma targeted reinventing PowerPoint because it was widely used and widely disliked
- •From 2020–2022, the team mostly built prototypes; a 2022 launch brought thousands of users but only modest activity
- •AI became the new “why now,” leading to a March 2023 launch that drove the first real PMF
Diagnosing the bottleneck: low activation revealed the blank page problem
Brett presses on why the blank page problem wasn’t already solved by incumbents, and Jon explains there was no technological alternative besides templates. Jon shares that once Gamma was in market, the team saw a glaring activation problem and realized the blank page was the obstacle.
- •Jon argues the blank page is a real problem in PowerPoint/Slides; users just lacked alternatives
- •Templates were the closest historical solution; generative AI created a new way through the problem
- •Jon points to “appallingly low” activation (around 5–10%) as the wake-up call
- •The team designed the launch around overcoming the blank page as runway dwindled
When onboarding becomes the product: eliminating hours of work and making creation “fun”
Jon explains that what looked like an onboarding fix turned out to be the real job-to-be-done for users: escaping the dread and workload of starting a presentation. He also recounts how internal dogfooding revealed an unexpected delight loop—generation felt playful and addictive.
- •Jon says fixing onboarding improved satisfaction because it solved the user’s real problem, not just Gamma’s funnel
- •He breaks down the workload of building a presentation and claims Gamma removes most of that time cost
- •Dogfooding sparked insight: generating presentations became “fun,” like a game or slot-machine reveal
- •The team didn’t initially expect onboarding work to unlock the full value proposition
Horizontal vs. vertical focus: resisting the investor playbook and picking a broad persona
Brett asks whether Gamma built for a narrow niche or a general audience, and Jon explains the deliberate bet on a horizontal product. He describes the fundraising tension this created and the logic behind choosing “external presenting” as a broad starting persona.
- •Jon says Gamma consistently chose a horizontal product strategy over a narrow vertical
- •He cites examples of broad tools the team admired and used as reference points
- •Gamma oriented toward external-presenting roles (sales/marketing/consulting) due to virality and willingness to pay
- •Jon notes investors repeatedly challenged this choice, creating ongoing friction
Early user research and hand-built prototyping before AI coding
Jon details how the company began with extensive qualitative research and rapid prototyping. He describes boiling interview findings into core problems and building many prototypes weekly, largely through dogfooding because early versions weren’t ready for customers’ high-stakes work.
- •Jon says the team started with 100 user interviews about the last time people made a presentation
- •The interviews revealed consistent pain across roles: starting, judgment about visuals, and formatting effort
- •The team distilled problems into formatting, structure, and content, then prototyped rapidly
- •Because presentations are high-stakes, the team often tested by making decks themselves
Why weak early fit became an advantage: rebuilding fast when AI primitives arrived
Brett asks whether weak early PMF helped Gamma pivot, and Jon strongly agrees. He argues the team had built useful non-AI infrastructure but wasn’t too attached to it, enabling a faster reinvention when AI became viable—while still not being just a “wrapper.”
- •Jon says having built foundational infrastructure helped, but weak fit made it easier to throw parts away
- •He contrasts Gamma with teams starting from scratch without product foundations
- •Jon notes the tension of evolving quickly now while serving a very large existing user base
- •Brett and Jon transition into how Gamma revisits earlier design bets as AI improves
Undoing old decisions: guardrails that helped in 2023 now hold back smarter models
Jon explains that Gamma’s early success required guardrails because models were weaker, but those constraints now limit what newer models can do. The company is shifting from adding features to removing constraints, while still keeping the user experience simple and predictable.
- •Jon says early LLM limitations drove Gamma to add guardrails to protect quality
- •He now believes those guardrails are “hamstringing” creativity and capability as models improve
- •The team is trying to “throw things out” and loosen requirements so AI can do more
- •They still must preserve a simple UX and avoid constantly adding new controls users must learn
A key mistake: shipping PowerPoint export too late and prioritizing “moat” over adoption
Jon identifies a major hindsight lesson: the team overestimated how much users would adopt Gamma’s new format and underestimated the need to interoperate with PowerPoint. He describes delaying PowerPoint export by about a year due to excitement about Gamma’s proprietary format and a focus on durability over adoption.
- •Jon says Gamma built a new responsive format between presentation/document/webpage
- •He calls out “drinking our own Kool-Aid” about users staying in the new format
- •Users kept exporting to PowerPoint because that’s where teams collaborate; Gamma shipped export too late
- •Jon frames the lesson as winning adoption first, then building stickiness and defensibility
Presentations as a hard category: episodic usage, bundling, and why incumbents stagnated
Jon explains why presentations have historically been a startup graveyard: usage is episodic and incumbents are bundled into suites. He argues incumbents have little incentive to innovate in slides, and that taste/creativity makes both product building and evaluation much harder than more objective domains.
- •Jon calls episodic usage and suite bundling major reasons startups fail in presentations
- •He says PowerPoint has remained dominant since its launch, with few serious challengers
- •He argues suite purchasing decisions rarely hinge on slides features, reducing incumbent focus
- •Jon highlights taste and creativity as core challenges, making “good” harder to define and measure
Evals for taste: using user data, A/B tests, and multiple models to learn preferences
Brett asks how Gamma handles evaluation when quality is subjective, and Jon explains their approach centers on extracting taste signals rather than a single objective score. He describes leveraging a large user base, testing models against each other, and using multiple providers to compare “feel” and performance.
- •Jon says they treat evals as a taste-and-preference problem, not purely objective scoring
- •Gamma learns from cloud-based visibility into what users make and how they prompt
- •They run A/B tests across models and can mix providers rather than being tied to one
- •Jon notes different image models have different textures, which Gamma can match to use cases
Building for fast model progress: temporary scaffolding, durable foundations, and fewer knobs
Jon describes an in-between engineering mindset: systems must scale to large usage but may only live 6–12 months before being replaced. He separates what must be durable (security, storage, rendering, preserving user artifacts) from what should change quickly (generation/edit mechanics), and argues for simplifying controls so advancing models don’t obsolete complex UI.
- •Jon describes a “scaffolding” approach: build supports, then remove them as models improve
- •He emphasizes feature flagging and designing with the expectation components will be thrown away
- •Durable commitments include privacy/security and preserving users’ existing presentations unchanged
- •He argues for fewer knobs and more reusable building blocks so AI can fill in capability over time
Monetizing an AI prosumer product: higher willingness to pay, AI tourism, and tiering
Jon explains that monetization in AI is a moving target and required ongoing adaptation. He argues AI products can monetize more easily due to clearer time-saved value, but introduces “AI tourism” churn—leading Gamma to tier pricing and distinguish episodic users from sticky ones.
- •Jon says no single pricing model would have worked unchanged from 2023 through later stages
- •He claims AI tools have higher willingness to pay because users map time saved to dollars spent
- •He describes “AI tourism”: users try many AI tools, so initial conversion doesn’t equal commitment
- •Gamma uses multiple tiers (entry, pro, ultra) plus team/business plans to match different usage patterns
The rapid leap into enterprise: demand pulled Gamma upmarket faster than planned
Jon says he expected a gradual climb from individual to SMB to enterprise, but AI-era competition and inbound demand forced a faster shift. He describes being initially reluctant due to the heavy lift of sales, compliance, and enterprise requirements—then building that motion while trying not to abandon the prosumer base.
- •Jon says enterprise expansion was not incremental; it became a rapid sprint
- •He attributes acceleration to both competitive pace and unexpected inbound from very large companies
- •He references prior experience seeing how difficult enterprise sales, compliance, and tooling can be
- •Gamma aimed to layer enterprise on top of a healthy prosumer business rather than replace it
Product vs. distribution, PMF signals, and the founder’s emotional journey
Jon argues product still matters more than distribution, claiming Gamma’s distribution inflected only after the AI product created built-in sharing and virality. He shares the quantitative PMF signal that matched the “rock downhill” feeling—signups per day inflecting upward—then closes with reflections on craft, speed, luck, and the ongoing intensity of founder life.
- •Jon disputes “distribution over product,” saying virality emerged after product improvements, not before
- •He points to signups-per-day inflection as the clearest PMF signal, describing rapid growth after the AI launch
- •He explains the cultural tension between craftsmanship/consensus and the need for faster iteration and willingness to sunset work
- •Jon reflects on emotional phases: early freedom, runway dread, then relentless “chasing mode,” plus his view that luck and execution both matter

Publisher timestamps

Get more out of YouTube videos.

Read along on your phone, search it beside the video, or ask your favorite AI app what was said.