Raycast founder Thomas Paul Mann - quality, YC and AI
Raycast co-founder Thomas Paul Mann explains how a faster Spotlight replacement grew into an extensible productivity layer — and why quality comes from dogfooding, narrow scope, direct feedback and software that makes people feel good.
Thomas Paul Mann co-founded Raycast in 2020 after years of wishing that the tools outside their IDEs felt as fast and keyboard-friendly as the tools inside them.
He traces Raycast from a turbulent Y Combinator batch to more than 2,000 extensions, then unpacks the operating habits behind its polish, its developer-led brand and its ambition to become an AI layer across the desktop.
The Q&A below is experimental — extracted from the episode transcript with AI. If you spot a mistake, let us know at jack@scalingdevtools.com.
Why did you start Raycast?
My co-founder and I were productivity nerds who increasingly had to leave our IDEs for slower, less keyboard-friendly tools. Command palettes were spreading from editors into products such as Superhuman and Linear, while our useful shell scripts remained inaccessible and visually limited. We imagined one global, fast command palette connected to every tool: a drop-in replacement for Spotlight, but ten times more useful to technical professionals.
What did Raycast learn from its earliest features?
We began with application and file search, then features such as clipboard history. The bigger insight came from interviewing engineers at YC alumni companies: people spent all day in Jira but disliked its everyday experience. Our first extension handled the common actions — creating, finding and updating issues — while leaving uncommon, complex work in Jira. That 80/20 model expanded into a store with more than 2,000 extensions.
How do you make software feel high quality?
Crash rates matter, but software can be stable without feeling well made. Raycast combines speed, an intelligent cache, careful visual design and a minimal interface with intense internal use. Everyone receives internal builds and lives with a feature for weeks before release, so visual glitches and workflow paper cuts surface before customers see them. Quality is the accumulated result of those small corrections.
How do you balance polish with shipping quickly?
We ship every two weeks and try to include a highlighted feature, which creates a real deadline. The team starts broad, builds enough to learn, then reverses the question: given the release date, which details are essential? We would rather ship a smaller first version at high quality than overbuild speculative functionality, then follow quickly with a second iteration informed by real use.
How does customer feedback reach the people building Raycast?
A Send Feedback command has existed since day one, and we reply to everything. Early users joined onboarding calls and sometimes received a fix within half an hour. As volume grew, an on-call rotation and then a support team handled replies, but feedback still flowed into Slack so engineers could see reactions to their own work. The goal is to keep the signal as close as possible to the person who can act on it.
Why do developers advocate for Raycast so strongly?
Developers prefer discovery to being sold, so we made choices we wanted as users: no account when functionality does not require one, anonymized analytics and a product that earns word of mouth. Paid marketing did not fit the brand or move the needle. Meetups, hackathons and open Q&A make the company feel human, while the product gives users an identity around fast, thoughtful work rather than another piece of utility software.
What does the Raycast brand stand for?
Productivity is less a number than the feeling of ending a day or week satisfied with what you accomplished. Raycast wants to help more people reach that flow state. The challenge is that it produces value through thousands of small actions rather than one dramatic outcome, so the brand must connect those actions to a larger feeling: well-crafted tools that help professionals do meaningful work.
How could AI change the way people use their computers?
Operating systems still make people translate an intention into applications, clicks and keystrokes. Raycast already sits close to that layer, with extensions and access to local tools, so it can let a person describe an outcome and perform the intermediate work. The beta is sometimes unreliable, but when it can generate an image, set a wallpaper or move research into Notion, it points toward an AI meta-layer connecting professional tools.
What becomes possible when users can extend that AI layer?
Natural language can turn a specialist workflow — such as an FFmpeg script for anonymizing research videos — into a small tool that colleagues can reuse without learning Bash. That returns to Raycast’s original purpose: make personal automation accessible, visual and shareable. Once an API is open, developers also build things we never anticipated, from highly personal extensions to running Doom inside Raycast.
Links & references
This episode is brought to you by WorkOS.