The post, written by Toss’s Developer Relations Manager, describes how the company approached AI transformation (AX) through a programme it called AI Surf Day. The author’s starting point is uneven adoption: since the previous year, people in product, design and staff roles as well as engineers had been working out how to use AI in their own areas, but some team members still felt distant from it, and people in non-development roles in particular felt the gap and grew anxious. The author notes that many companies are pursuing AX in different ways because there is no settled answer, and presents AI Surf Day as one of Toss’s own.
AI Surf Day turned every Friday from April to June into a day for experimenting freely with AI and applying it to work, with Monday to Thursday kept for regular duties. Its name and concept come from the line “You can’t stop the waves, but you can learn to surf”, attributed in the post to Jon Kabat-Zinn: the pace of AI cannot be held back, so the idea is to learn to ride it together. The author describes it as having become a cultural device for making Toss a company that works on the basis of AI, beyond simply getting everyone to use AI.
The programme had three fixed components — employee-run clubs, a case-sharing session and a network of evangelists — plus occasional one-off variations, one of which was a collaboration day with OpenAI that included a hackathon built around agentic workflows. The post closes by saying that the programme was scheduled to run until June, that whether it continues was still under discussion, and that the form matters less than what it left behind.
Key Points
- AI Surf Club let employees open and join free-form groups on AI-related topics; the post says about 200 clubs were created as the programme began. Examples it gives include a study group on AI anti-patterns, where people already using AI shared their failures and wrote them up as a guide; a “step 0” group for colleagues who had not yet managed to install an agent tool; and a customer protection team that, according to the post, built a portal for monitoring external complaints with AI within a month, alongside automations for drafting complaint replies and classifying complaints.
- A business marketing team is described as dividing AI work into Builder, Curator, Operator and Scouter roles through an internal AI Native Workshop, with the stated aim of leaving each person’s output in a form colleagues could copy or apply to their own work, rather than stopping at “let’s use AI well”.
- AI Surf Weekly shared internal best cases, lessons learned and recent AI insights. Before the programme the author had surveyed employees’ AI use cases and introduced people with similar needs across different organizations to each other; from that the author concluded that introducing good users and good cases lets further good cases emerge on their own, and chose to share ideas rather than tool-by-tool skill tips.
- The post argues that what it means to use AI well differs by organization and role, and that for an organization it depends not on how well individuals handle AI tools but on redesigning the team’s existing workflow on the basis of AI and solving the problems that process surfaces.
- On that reasoning the AI Surf Evangelist scheme selected people who were most committed to introducing and spreading AI in their teams, rather than the most technically fluent. Nominations, each with a short recommendation, were collected in a public channel during the first week of April across all Toss affiliates, and 142 evangelists were chosen. Their missions over three months were to report good AI use cases to an evangelist channel, hold at least one offline meetup or workshop for their organization, and spread good cases and insights to their teammates; the Culture team supported the workshops with templates and facilitation, many of them on the theme of redesigning the team’s work on the basis of AI.
- The OpenAI collaboration day on May 15 opened with hands-on sessions — one for developers on a Codex-centred development workflow (OpenAI Codex) and one for non-developers on automating work with ChatGPT Agent — followed by a two-and-a-half-hour mini hackathon with Agentic Architect and Agentic Operator tracks.
- The winning Architect entry, as the post describes it, had Codex operate the iOS Simulator itself to verify features: asked to build a comment feature, it implemented the feature, logged in, posted a comment and checked that it worked, looping through plan, code, build, simulator verification, fix and repeat until done, and leaving a video as evidence. The winning Operator entry had an AI agent make a first-pass classification of newly created product data against a predefined scheme and message a person in charge, who confirms or rejects the result in an admin tool — described as turning repetitive work into a reusable agentic workflow.
Context
The post is a firsthand account from the organization that ran the programme, written by the person who designed parts of it, and it reports participants’ reactions through quotations rather than measured outcomes. Its own caveats are explicit: Toss is not proceeding with an answer sheet but is experimenting and working things out together, the author cannot say whether the company is surfing well, and the change may turn out to be abrupt or gradual. The author’s closing judgment is that what AI Surf Day left was a time slot set aside for learning and an atmosphere in which people share “I tried this and this happened” without hesitation, and that culture and people matter more than programmes or systems.