He Built the World's #1 Open-Source Coding Agent

Predictions and Expectations:

• Jay Poldinger expects that "the market for intelligence has not existed before" and that "everybody should be thinking in a positive, some very high mentality" regarding the sector.
• Jay Poldinger predicts that models are becoming "commoditized utilities" and that the industry will see model labs "picking off niches and chunks" by specializing in specific areas or characteristics.
• Jay Poldinger believes that if the AI industry operates as a "positive sum" growth market, it is in the interest of model labs to want "OpenCode to grow and succeed" because they will become "huge customers" for them.
• Jay Poldinger predicts that open source models will become "good enough for real work," which triggered a wave of users to try OpenCode as an alternative to frontier models.
• Jay Poldinger fears that the "chasm" of high token costs for AI usage is "very hard for a lot of people a lot of companies to cross," necessitating subsidies to help users reach the "whales" stage.
• Jay Poldinger expects that enterprises will increasingly use OpenCode not just for developers but to allow "non-technical people" to utilize coding agents as part of their "core loop."
• Jay Poldinger believes that being intentional about product design to ensure "everybody uses every single day" is what resonates with users rather than just "cheaper tokens."

Timelines and Milestones:

• Jay Poldinger plans to launch the "subscription product" around "end of September, early October last year," noting it took "about eight months" to reach "$40 million or so" in inference revenue.
• Jay Poldinger reports that the subscription product launched "early March" and grew to "160 000 monthly subscribers" by the time of the interview.
• Jay Poldinger cites a specific inflection point in "February of this year" where for a "four week span" users utilized Kimi models more than Anthropic models (Sonnet plus Opus).
• Jay Poldinger notes that by "August, September or so" last year, the first crop of open-source models emerged, followed by a gap where models were "six months behind" frontier capabilities.
• Jay Poldinger references the "beginning of the year" as the starting point for a "20x increase" in monthly active users, reaching "13 million" by June.
• Jay Poldinger mentions that Cloud Code launched in "February or so of 2025," which was a catalyst for OpenCode's development as a terminal UI alternative.
• Jay Poldinger states the company incorporated in "2010" and that the legal entity is "almost 20" years old, with the current product trajectory building on "16 years" of existence.
• Jay Poldinger estimates that the company is "10 years" into a journey to reach the current scale, having gone from zero to "30 million in eight months."

Technology and Product Direction:

• Jay Poldinger plans to allow users to "switch over to one of these very cheap models" like DeepSeek Flash when approaching daily or weekly limits to extend usage.
• Jay Poldinger expects that users will optimize for "token budgeting" by utilizing cheaper models for specific tasks, such as using GLM 5.0 for "front-end design" when it exhibits superior performance in that area.
• Jay Poldinger predicts a shift in the enterprise world where organizations will seek to "limit access" to frontier models for specific internal groups to manage "token spend" more creatively.
• Jay Poldinger plans to maintain a "free tier" to drive the "magic moment" for new users, while offering a subscription plan for those doing "real work" to justify spending to "redo some of the processes" within their companies.
• Jay Poldinger intends to showcase "diversity" of models to create an environment where "competition ends up being good for the consumer" rather than locking users into specific vendors.
• Jay Poldinger describes the product architecture as having a "server" component (the agent loop) that can be "embedded separately from the UI," allowing for integrations like Ramp's Slack bot.
• Jay Poldinger believes that the "name OpenCode" was a deliberate choice to occupy the position of an "open alternative" to dominant players, aiming to become the "default" as quickly as possible.

Market and Industry Outlook:

• Jay Poldinger notes that the market for coding agents is "huge in developing countries" like Indonesia ("4% of your traffic") and Brazil ("5% of your traffic") where a "$200 a month Cloud Code subscription is like very expensive."
• Jay Poldinger predicts that "most people in the world still haven't experienced the magic of a coding agent" and that OpenCode's goal is to get "as many people in the world as possible" to have that "aha moment."
• Jay Poldinger observes a "vibe shift especially in the enterprise world towards like more token budgeting" and that the industry is moving away from ignoring costs toward "conscious with the tokens."
• Jay Poldinger expects that the "frontier models and the frontier labs" charging high prices per token makes it "hard for a lot of people across the globe to have that experience," creating an opportunity for open-source alternatives.
• Jay Poldinger believes that the "market is so large that it is hard to imagine people not, or model labs not picking off niches" and that "choices only improved over time" as evidenced by growth.
• Jay Poldinger reports that "17%" of traffic comes from China, noting that "a lot of these models are, are Chinese," and OpenCode allows users there to use models "being built in, their country."
• Jay Poldinger predicts that "whales" will emerge where a "certain percentage" of users cross the usage cost chasm and begin "spending the the crazy amounts," making the funnel viable.

Company Plans:

• Jay Poldinger plans to "take that [magic moment] to as many people in the world as possible" using the free tier to overcome the barrier of high frontier model costs.
• Jay Poldinger intends to build a "subscription plan that allows them to do real work" with open source models so users can "justify spending so much more" to change company processes.
• Jay Poldinger plans to leverage a "reasonably stable 24-hour GPU cycle" to achieve "pretty good utilization" and "competitive advantage" by serving a global user base.
• Jay Poldinger aims to maintain a product bar where "everybody uses every single day" and "holds that bar fairly high" to resonate with enterprise users who "bug" them to sign security agreements.
• Jay Poldinger plans to continue publishing data to show "actual usage data for each of the users" rather than just aggregated views, to highlight model performance realities.
• Jay Poldinger believes that the company's strategy involves "betting the field" rather than picking a specific model lab winner, as the "rest of the field is going to do well."
• Jay Poldinger expects that the "open source ecosystem" needs to work as a whole, with OpenCode acting as the "largest customer for most of them" to drive the machine's success.

Financial Guidance:

• Jay Poldinger projects that based on June's data extrapolated for the year, revenue from inference would be "around $31 to $33 million or so."
• Jay Poldinger projects that based on "last week's data, extrapolated it for the air [year]," revenue would be "around $38 to almost $40 million."
• Jay Poldinger reports that the subscription product "accounts for about 18 million of sort of the annualized revenue."
• Jay Poldinger notes that "160 000 monthly subscribers" are paying for the subscription plan.
• Jay Poldinger estimates the company is processing "around 7 trillion tokens per day," which is double OpenRouter's total of "around 6 trillion."
• Jay Poldinger cites an earlier data point of starting the year with "around $300 billion or so" in token volume.

Risks and Caveats:

• Jay Poldinger hedges the claim that "everybody should be thinking in a positive, some very high mentality" by framing it as a recommendation rather than a certainty.
• Jay Poldinger qualifies the growth data by noting that the "20x increase" is from "the beginning of the year" and that the "bump in monthly active users" can be tracked to specific market events.
• Jay Poldinger caveats the data interpretation by stating "we have sort of some theories around why" DeepSeek's token volume dipped when GLM emerged.
• Jay Poldinger warns that the "unit economics" rely on "volume discounts" from aggregating tokens, and without that volume, the cost structure changes.
• Jay Poldinger notes that while the "free tier" drives acquisition, it involves "eating the cost," similar to how "Anthropics and the opening eyes of the world" subsidize usage.
• Jay Poldinger observes that while some models like DeepSeek Flash are "cheap," the "token maxing is expensive" and represents a significant "chasm" for many users to cross.
• Jay Poldinger admits that the "name OpenCode" came "a little bit later" and was a deliberate choice made when the founders felt the market needed an "open alternative" to two dominant players.

Confidence and Disagreement:

• Jay Poldinger expresses confidence that "you really know you have product market fit when like enterprises are bugging you to sign the security agreement."
• Jay Poldinger states with certainty that "OpenCodes has been growing at an astounding rate this year."
• Jay Poldinger indicates a strong belief that the "market for intelligence has not existed before" and is "unprecedented."
• Jay Poldinger disagrees with the Twitter narrative that GLM is "taking over DeepSeek," citing data that shows DeepSeek still leads in token volume.
• Jay Poldinger notes that "it is a little bit hard to see" the top unique user counts on the market share graph but states the data shows "glm 5.2 at close to... 30k."
• Jay Poldinger speculates that the "meme" of Cloud Code banning OpenCode "inadvertently" put the two products on the "same sort of pedestal," equating them in users' eyes.