CHAPTERS
- 0:00 – 1:23
OpenCode’s mission: getting everyone to experience a coding agent
The hosts set the premise: coding agents are still new for much of the world, and OpenCode aims to deliver that “aha” moment broadly. They introduce OpenCode as an open-source, model-agnostic alternative to Claude Code that’s rapidly gaining adoption.
- •Coding agents are still underexperienced globally
- •OpenCode positioned as open-source and works with any model
- •Goal is broad access to the “magic” moment of agentic coding
- •Early hints of global adoption and enterprise pull
- 1:23 – 3:07
Explosive growth metrics: 20× MAUs, 7T tokens/day, and a fast-growing business
Jay shares the scale and pace of adoption: massive user growth, token throughput, and quickly ramping revenue. The team breaks down how subscriptions and inference contribute to annualized revenue, plus the size of the paying base.
- •~13M monthly active users and ~4.6M weekly actives
- •~20× growth since start of year
- •~7 trillion tokens processed per day (surpassing major aggregators)
- •Inference + subscription revenue tracking toward ~$40M annualized
- •~160k monthly subscribers; subscription portion ~18M annualized
- 3:07 – 5:40
Codex interoperability and the Anthropic clampdown that boosted awareness
They discuss how some Codex subscribers use OpenCode as their primary interface and how official support made that easier. Jay recounts the early-year controversy where Anthropic tried to block OpenCode-related usage, which inadvertently elevated OpenCode’s profile.
- •Codex officially supports using its subscription through OpenCode
- •Anthropic attempted to block usage tied to Claude Code subscriptions
- •Blocking behavior reportedly triggered by the word “OpenCode” in prompts
- •Controversy created an inflection point by putting OpenCode and Claude Code “on the same pedestal”
- •More people discovered OpenCode because of the clampdown narrative
- 5:40 – 8:54
Why open models made OpenCode’s global vision viable
Jay explains the product’s global framing: frontier tokens are expensive, so OpenCode helps more people access agentic coding. The conversation tracks how the quality gap narrowed and open models became “good enough” for real work, enabling OpenCode’s subscription strategy.
- •Frontier model pricing limits access worldwide
- •Initial use case: bring-your-own Claude subscription into OpenCode
- •By late summer/fall, open models felt ~6 months behind but catching up
- •Waves of adoption tracked new open-model releases
- •A tipping point: open models becoming viable for serious work
- 8:54 – 11:37
Model usage insights: token volume vs unique users (what developers actually pick)
They walk through OpenCode’s published usage dashboard and what it reveals beyond social-media hype. The team contrasts token volume and unique-user adoption, noting which models dominate and why perceptions can differ from reality.
- •OpenCode publishes usage stats at opencode.ai/data
- •Token-volume leaders include DeepSeek variants and GLM
- •Twitter hype doesn’t always match real usage distribution
- •Unique-user counts show close competition among top models
- •OpenCode’s view is more user-granular than aggregated providers
- 11:37 – 13:28
Token budgeting behavior: cost, speed, and task specialization drive model choice
They unpack why cheap models can dominate: users ration limits, switch models late in a cycle, and optimize for speed. Jay also notes “fit-for-task” behavior—certain models gain adoption for strengths like front-end work.
- •Many users switch to cheaper models when nearing usage limits
- •Throughput (tokens/sec) can matter as much as quality
- •Kimi adoption was partly driven by perceived real-time speed
- •Model selection can be task-dependent (e.g., GLM for front-end design)
- •Enterprise world increasingly cares about token budgeting too
- 13:28 – 15:24
Geographic breakout: huge adoption in China and developing markets, plus US growth
OpenCode’s subscription plan was built for global affordability, and the geo chart reflects that with significant usage in China and developing countries. Surprisingly, US usage is also strong—suggesting broader token-consciousness and interest in trying emerging models.
- •China is the largest share (~17%) in the discussed snapshot
- •Developing markets (e.g., Indonesia, Brazil, Vietnam) show meaningful traffic
- •High-priced subscriptions (e.g., $200/month) are prohibitive in many regions
- •US growth is notable despite expectations that US users ‘just spend’
- •Model experimentation (e.g., GLM spikes) drives subscription trials
- 15:24 – 16:33
Why enterprises pick OpenCode: neutrality, flexibility, and daily-driver product quality
Beyond cost, enterprises adopt OpenCode to avoid lock-in to a single model or interface. Jay notes strong Fortune 500 traction and emphasizes intentional product decisions aimed at being something teams use every day.
- •Enterprises want model/harness choice and future flexibility
- •Reports of thousands of employees using OpenCode inside large companies
- •“Forward” Fortune 500 usage is significant but often confidential
- •Focus on being a high-quality daily tool, not just a cheaper option
- •Open-source neutrality helps procurement and long-term adoption
- 16:33 – 19:59
AI token economics: the new CAC, subsidies, and whales as the funnel unlock
They discuss how token spend replaces traditional ad-driven CAC: you must subsidize usage so users can reach expert proficiency with agents. Jay explains OpenCode’s approach—free tier for the aha moment, subscription for real work, and high-usage customers powering margins.
- •Learning to use coding agents well can be expensive in tokens
- •Frontier labs subsidize onboarding; power users (“whales”) justify the model
- •OpenCode uses free tier + affordable subscription to cross the chasm
- •Whales paying per token can drive strong margins at scale
- •Volume-based discounts on tokens translate into unit economics advantages
- 19:59 – 28:43
Enterprise deployment patterns: procurement pull, token controls, and embedded agents (Ramp example)
Jared probes what enterprises ask for after adoption: security questionnaires arrive after organic internal usage, indicating strong pull. Jay describes requests around token governance, visibility, and embedding the agent loop into other products—highlighted by Ramp’s Slack bot implementation.
- •Enterprises often reach out after employees already adopted it
- •Typical ask: security questionnaires and formal approval processes
- •Requests for organizational token limits and model access controls
- •Some ask for detailed activity visibility (privacy/product tradeoffs)
- •OpenCode has separable UI + agent-loop ‘server’ for embedding (e.g., Slack bot)
- 28:43 – 29:52
Scaling inference: global usage smooths GPU demand and improves utilization
As a major customer of open model inference, OpenCode rents GPUs and works with providers and model labs. Jay highlights a structural advantage: worldwide usage creates a more stable 24-hour demand curve, improving utilization and cost efficiency versus regionally concentrated products.
- •Compute strategy spans rented GPUs, inference providers, and model labs
- •Global peaks/troughs are flatter due to time zone diversity
- •More consistent utilization helps unit economics
- •Operational efficiency becomes a competitive advantage
- •OpenCode’s scale makes it a major inference driver for open models
- 29:52 – 34:18
Deliberate product strategy: ‘open alternative’ positioning, broad model support, and terminal-native UX
Jay explains the intentional choices behind OpenCode’s moat: becoming the default open alternative in a market with dominant closed players. They built extensive model/provider support (including Models.dev) and prioritized a ‘modern terminal’ experience for core developers.
- •Strategy: the market coalesces around an open alternative to dominant players
- •Name ‘OpenCode’ and position were chosen deliberately
- •Claimed 70+ models/providers early; required building Models.dev database
- •Terminal-native UI designed for developer ‘daily driver’ expectations
- •Deep roots in open-source community shaped design and adoption
- 34:18 – 41:13
A 16-year ‘overnight success’: long road through YC attempts, open source, and compounding experience
The conversation shifts to Jay’s backstory: starting in 2006–2007, incorporating in 2010, and applying to YC many times before acceptance in 2021. Jay frames OpenCode’s success as the sum of prior waves, products, and skills—now unified into one go-to-market spanning consumer to enterprise.
- •Origins in college era; early YC interviews in the PG days
- •Same legal entity persisted across many idea iterations
- •Multiple YC applications/interviews before getting in (accepted in 2021)
- •2021 product: serverless platform/framework; first major open-source push
- •Building in public became identity; past experience now applies across segments
- 41:13 – 44:26
Why Jay never quit—and how to try OpenCode today
Jay attributes perseverance to stubbornness, steady learning, and incremental progress—even through lean periods. The episode closes with practical advice: OpenCode makes it easy to try new models via a model picker, especially for developers who want to experiment beyond closed ecosystems.
- •Persistence driven by visible progress and skill accumulation
- •Acknowledges the path was risky (living with parents at times)
- •Motivation: learning product, marketing, and positioning end-to-end
- •Recommendation: use OpenCode to quickly try new open models
- •Closing reflections: being prepared enables ‘lightning in a bottle’ moments
