Augment Code: A Pure Storage Founding Engineer and a Google AI Researcher Raised $252 Million in Stealth, Then Learned How Hard AI Coding Pricing Is

The story

Igor Ostrovsky built Augment as the tool he wanted to use himself. "Software engineering remains a difficult and all-too-often tedious and frustrating job, particularly at scale," he told TechCrunch when the company came out of stealth. "AI can improve software quality, team productivity and help restore the joy of programming." His background was in systems rather than machine learning. Ostrovsky spent nearly seven years at Microsoft, where he worked on components of Midori, a next-generation operating system the company never released, and then joined the flash storage company Pure Storage as a founding engineer. Augment's own announcement later described him as Pure Storage's former chief architect.

His co-founder supplied the AI expertise. In 2022, Ostrovsky and Guy Gur-Ari, previously an AI research scientist at Google, teamed up to build Augment's first product, an MVP. To fill out the leadership, they recruited Scott Dietzen, the former CEO of Pure Storage, and Dion Almaer, a former Google engineering director and VP of engineering at Shopify. The company said its engineers came from Google, Meta, Nvidia, Microsoft, Databricks, Snowflake, VMware and Pure Storage.

Their diagnosis of the market was that existing AI coding assistants were too shallow for real enterprise work. "Most companies are dissatisfied with the programs they produce and consume; software is too often fragile, complex and expensive to maintain with development teams bogged down with long backlogs for feature requests, bug fixes, security patches, integration requests, migrations and upgrades," Ostrovsky said. The company's launch announcement argued that current assistants did not adequately understand a programmer's intent, improve software quality, help team productivity or properly protect intellectual property.

Augment stayed hidden until it had raised a great deal of money. On April 24, 2024, it emerged from stealth announcing a $227 million Series B at a $977 million post-money valuation, with Sutter Hill Ventures, Index Ventures, Innovation Endeavors, Lightspeed Venture Partners and Meritech Capital; together with a $25 million Series A led by Sutter Hill, total funding reached $252 million. Former Google CEO Eric Schmidt invested as well. "Software remains far too expensive and painful to develop," Schmidt said in the announcement. Sutter Hill's Mike Speiser called the team "as good as any team we have ever helped put together."

Even with all that money, the founders said little about the product. TechCrunch described Augment as "a strangely hush-hush operation," noting that Ostrovsky would only say the company used fine-tuned "industry-leading" open models and would make money through standard subscriptions, with pricing to be revealed closer to general availability. He said "hundreds" of developers across "dozens" of companies were using it in early access, and that the funding "provides many years of runway." The competition was already intense: GitHub Copilot had more than 1.3 million paying individual users and 50,000 enterprise customers, and startups such as Magic, Tabnine and Cognition were all chasing the same buyers. TechCrunch also pointed out that compute costs had forced the coding assistant startup Kite to shut down in December 2022, and that even Copilot was reported to lose money per user. That warning would prove relevant.

Augment's differentiator became what it called its Context Engine, technology meant to give each AI request the right information from a large codebase. On April 2, 2025, Gur-Ari announced Augment Agent, which he said could deliver high-quality code "from a brand new app to a monorepo with over 100k files." He wrote that context "is what sets us apart from our competition and what our users love," and pointed to a first-place spot on the SWE-bench Verified leaderboard. The agent introduced Memories, which persist across conversations to match a developer's code style, supported the MCP protocol for connecting tools, and offered a context capacity of up to 200,000 tokens, which the company described as twice that of comparable tools.

Then came the problem every AI coding company eventually faces: the cost of running models for heavy users. Augment launched the agent with temporary unlimited pricing, and on May 7, 2025 switched to plans based on "user messages," charging for each successfully processed request a user sent. The company said it chose that metric after users told it they did not want to be billed for tool calls and actions they could not control, nor to "do the math of what a credit is worth." The Register reported that the earlier plans had offered unlimited chats and completions for $30 or $60 per user per month, and that the May plans ranged from 50 free messages to 4,500 messages for $250, prompting one developer to complain it "now costs more than Cursor and Windsurf combined."

Five months later, Augment reversed itself. On October 6, 2025, CEO Matt McClernan announced that from October 20 every plan would come with a pool of credits instead of messages. He wrote that the message model treated a request to edit a CSS class the same as a request to refactor a mobile payment view and add Apple Pay, and that "the user message model also isn't sustainable for Augment Code as a business." As an example, he said one user on the $250 Max plan had issued 335 requests per hour, every hour, for 30 days, and was approaching $15,000 per month in cost to the company. "This change will increase costs for some Augment Code users," he wrote, adding that "our heaviest users will likely feel the change in price the most."

The reaction was sharp. The Register reported that users calculated the new model would be more than ten times as expensive for them; one said an email showed their 31 messages in a week corresponded to 40,982 credits under the new system. "I'm out. It was good while it lasted," that user wrote. Another accused the company of using early users to refine its system and then pricing them out. The Register framed it as the second major price increase in six months.

Augment's founders built credibility with a strong systems team and an enterprise focus on large codebases, and they raised enough money to take their time. Their pricing saga, though, captures a structural problem across AI developer tools: flat or message-based plans attract the heaviest users, and model costs scale with those users' ambition rather than with what they pay. McClernan's explanation was candid and arguably correct, but changing the rules twice in six months damaged trust with exactly the developers who had championed the product. For founders in this market, the lesson is to design pricing that tracks real costs from the start, because walking back generous plans later is far more painful.