Closing in on $50M with an AI coding agent by building for next
year
Igor Ostrovsky left his job in 2021 to dabble in AI. He quickly found a big opportunity and built for it. After four years of building — and only 6 months since public launch — Augment Code's AI coding agent is estimated to hit $50M+ within the year.
Here's Igor on how he built ahead of the curve. 👇
Lighting the fuse
I grew up in Slovakia in the late 1980s, hacking on a Czechoslovak PMD-85 my mom brought home from work. When communism in Czechoslovakia collapsed in 1989, we could travel across the border to Vienna, and my parents bought me a Commodore 64. Those machines — and the BASIC books my parents found for me — lit the fuse for a lifelong obsession with computers.
I taught myself to program and also worked part-time as a laborer at a fish plant in high school and early college. I studied computer science at the University of British Columbia, performed well in programming competitions, and ultimately spent eight years building distributed-storage systems at Pure Storage.
In late 2021, I left Pure to explore large-language models. The leap felt risky, but my wife backed the adventure: “Give it a chance — you can always go back to storage,” she said.
GPT-3 had hinted that language models could reason about code. I prototyped on a single GPU, joined open-source research groups, and saw flashes of something useful — models that could edit or explain unfamiliar files. The conviction hardened when our early context-aware completions produced answers no other tool could match.
That exploration became Augment Code, and a year later, real customers felt that same “aha” the first time Augment surfaced the exact function buried 10,000 lines away. Today, we're an AI coding platform that helps professional engineers ship real software inside huge, fast-moving repositories.
We’re venture-backed and growing extremely fast, but we intentionally keep absolute revenue figures private. Our board has hinted at revenue goals for the year north of $50M — I’ll neither confirm nor deny.
Shipping two big rocks
Our v0 focused on the workflows engineers perform every minute: writing, editing and navigating code inside their IDEs. We took just over a year to ship a private alpha dominated by two “big rocks”:
Our Context Engine – a real-time index that handles multi-branch updates, millisecond latency, and codebases with hundreds-of-thousands of files.
A custom code-completion model – high-quality code LLMs were scarce in 2022, so we trained our own before great OSS checkpoints existed.
We resisted the temptation to ship UI polish first; the hard AI and systems work had to come early so everything else could scale later.
Secure and scalable stacks
Internally we blend research and systems engineering: custom retrieval pipelines, latency-critical inference services, and IDE extensions in VS Code and JetBrains.
The stack has evolved with each wave of model capability, but the design principles stay constant — full-repo context at interactive latency and architectures that can pivot quickly when a new model (or use-case) lands.
Our architecture emphasizes security, scalability, and deep codebase understanding. Here’s a non-exhaustive list of what we use:
Languages: Go, Python, Rust
Architecture: Microservices on Kubernetes with gRPC inter-service communication
Client Applications:
VS Code Extension: TypeScript (clients/vscode/)
JetBrains Plugin: For IntelliJ and other JetBrains IDEs
Vim/NeoVim: Native integration
Frontend and UI:
Web: React with TypeScript
Styling: Tailwind CSS
Build Tools: Vite, Bazel
Components: Radix UI, custom design system
Data and Analytics
Databases: BigTable for analytics and logging
Monitoring: OpenTelemetry, Prometheus metrics
Build for next year
Every software epoch has its “platform shifts,” but AI moves faster than anything we’ve seen before. Mobile and cloud evolved on a five-year cadence; with large-language models we now see one or two genuine waves per year—retrieval-augmented chat in 2023, interactive agents in 2024, something new already brewing for 2025.
The job, therefore, is to ship for the current wave while secretly building foundations for the next one. That means modular retrieval layers, model abstraction seams, and constant vigilance about whatever breakthrough paper lands on arXiv next week.
A normal startup already demands hustle; an AI startup piles on a half-life measured in quarters. I look for engineers (and managers) who are comfortable making good decisions before the ink on yesterday’s paradigm is dry. Concretely, that shows up as:
Bias for outcome-driven agency (“We have to ship Remote Agents—what walls do we run through first?”)
Willingness to trade old certainties for rapid learning loops
Instinct to align, then self-propel — ability to deeply focus without getting distracter
Not everyone enjoys that tempo, and that’s okay, but we look for future teammates who have deep focus and a desire to move really fast.
Breaking through in a crowded market
Our focus is on building the best product available on the market for professional software engineers.
We make it really easy for people to try Augment Code. There is so much hype in AI, that we want teams to try Augment so they can see the difference our Context Engine makes for large codebases. It’s a difference you feel immediately when you compare to vibe coding tools.
The challenge is breaking through in a crowded market. We try to do that by partnering with software engineering voices professional developers trust and listen to, investing in technical content that explains what's under the hood, and promoting Augment in developer-heavy spaces.
A "try first" model
Pricing for AI tools today is generally complicated and confusing: A lot of companies make you calculate tokens, tool calls, and have model-based pricing. We’re trying to keep it simple.
We price on successful user messages sent, basically: requests you make to Augment Code whether those are chat or Agent. Augment has a 200k token context window and uses Claude 4 under the hood for Agent — no model picker, no switching, just the best code model.
Hands-on trial. Developers start with a 14-day unlimited free trial—no sales call required.
Self-serve onboarding. The IDE extension installs in minutes, so teams can evaluate real workflows on their own code.
Enterprise requirements. All the things professional teams need: SOC 2 Type II, ISO 42001 alignment, customer-managed keys, and detailed security documentation.
Assisted purchase. When an organization is ready to deploy at scale, our team steps in to handle procurement, security reviews, and training.
This combination — try first, buy when ready — aligns with how professional software teams adopt new tools while still meeting their security and compliance standards.
Bet forward
Here's my advice:
Understand the technology, know the customer, know the market—deeply. Combine those three lenses before choosing direction.
Bet forward. In AI, yesterday’s “cutting-edge” will commoditize in months; plan for the model you expect to exist a year from now.
Look at primary sources. We get hands-on with primary sources — reading papers, poking at model code, training toy checkpoints. It beats any second-hand summary. Joining OSS research chats and getting hands-on building accelerated my learning curve more than any book could.
What's next?
Professionally, the mission is clear: Automate large chunks of the software-development lifecycle with trustworthy agents.
Personally, my ambitions are clear: Build Augment, be present for my family, and stay healthy enough to keep doing both.
You can follow along on X and LinkedIn. And check out Augment!
Leave a Comment
Closing in on $50M with an AI coding agent shows the massive potential of building forward-thinking tech. Focusing on next year’s needs today ensures you're not just ahead—you're defining the future of software development.
Really enjoyed reading Igor’s story. What struck me most is how much it resonates with what we see in IP and legal tech.
When I started iPNOTE, I also had that moment of “manual chaos” — endless emails with local IP firms, unpredictable costs, missed deadlines. It felt like there had to be a better way. Just like Igor built Augment Code by going deep into code context, we built iPNOTE by going deep into the workflows of IP lawyers and companies managing global portfolios.
One parallel I love: Igor talks about building ahead of the curve. In IP, the same principle applies. Filing trademarks or patents is not just about the rules today — it’s about anticipating the next wave of regulation, automation, or AI adoption. If you don’t “bet forward” (to borrow Igor’s words), you’ll always be catching up.
It’s inspiring to see Augment Code scale so fast. Different industry, same core lesson: true defensibility comes from pairing domain expertise with timing and the courage to move before the market is ready.
This was a fantastic read, thanks for sharing the inside story, Igor. What stood out to me most was your point about “building for the current wave while secretly preparing for the next one.” That’s the exact tension we’ve felt in voice AI: shipping something useful today while laying down the rails for whatever tomorrow’s models will demand.
Also really resonate with the idea of resisting UI polish until the core AI and systems work is solid. It’s tempting to chase surface wins, but real breakthroughs seem to come from those invisible foundations.
Appreciate the transparency here, it’s not often you get to see both the technical stack and the philosophy driving it. Excited to watch Augment keep pushing the edge
Igor’s journey with Augment Code is a testament to bold conviction, technical depth, and timing. Leaving a stable role to explore LLMs in 2021 shows the kind of risk-taking that drives real innovation — especially when grounded in years of systems engineering. What stands out is not just the early technical insight, but the discipline to build for scalability, developer trust, and real-world workflows. In a crowded market, shipping a tool that professionals choose to adopt is rare. Igor’s story reminds us: breakthrough products don’t just ride the wave — they shape it. Hats off to a thoughtful, forward bet.
Just wow
What an incredible story! Proves that solving a real problem with great execution beats being first to market every time. Inspiring stuff, Igor Ostrovsky!
Incredible execution and foresight. Building for the next wave while delivering value now is what separates lasting startups from hype cycles. Huge respect for the deep technical foundation and focus on real workflows.
Reading this, it’s clear how essential it is to build with a long-term vision, especially in AI where change is rapid. The ‘try first’ approach seems like a great way to introduce developers to Augment Code. Looking forward to seeing how this platform continues to grow
impressive!
Incredible growth — going from concept to $50M+ projections in such a short time shows how fast AI-native tools can scale when they solve a critical workflow.
Curious — with AI assistants like ChatGPT and Claude increasingly shaping developer decisions, do you see visibility in AI-driven recommendations becoming just as vital for adoption as traditional marketing?
Investing in AI coding agents today will position companies for success in the future, with scalable automation and innovation potential to drive revenues close to $50M by the end of the decade.
I used to build growth strategy decks at McKinsey.
Now I’ve turned that expertise into a tool anyone can use, in 30 seconds, for $9.99.
It’s called Biz Boom and it’s designed for small business owners who are doing the work, but need clearer direction on what actually moves the needle,
You get,
• A niche-specific growth plan
• Offer and messaging ideas
• Step-by-step actions you can apply today
• All tailored to your business, instantly
No fluff, just what to do, and why,
Check this out before price goes up, you’ll be surprised how good it is for $9.99.
¡Qué historia tan inspiradora!
Desde un PMD-85 en la era comunista hasta una plataforma de IA con ambiciones de $50M… eso es una evolución épica 🚀
Me encantó cómo mezclaste curiosidad autodidacta, competencias técnicas y visión para detectar el potencial de los LLMs cuando todavía era algo experimental. Lo de “encontrar una función enterrada 10,000 líneas más abajo” me hizo sonreír — es justo el tipo de dolor que cualquier desarrollador ha sentido y agradecería resolver.
Estoy trabajando en un proyecto más pequeño con IA, pero leer cosas así realmente motiva. Mucho éxito con Augment Code. ¡Espero ver más avances pronto!
it's a massive, if it's succeed then it's going to be multi billion dollar org
🤯 $50M with a coding agent!! massive respect. The idea of “building for next year” really hit hard. Too many of us are building for last month’s trend instead of skating to where the puck will be.
I’m working on something much smaller (and much weirder), an AI character called Angry John, imagine a foul-mouthed drill sergeant with PTSD answering dumb internet questions 😅. Totally satirical, but we’re betting hard on voice AI, character IP, and viral entertainment.
Curious: Did you ever have moments early on where people thought your direction was too “out there”? How did you validate while still thinking long-term?
"This is such a powerful example of building with deep conviction and long-term vision. Love the focus on real workflows, full-repo context, and keeping things dev-friendly with the 'try first' model. As someone working on AI tools for game development, this resonates big time — especially the 'build for next year' mindset!"
With an AI coding agent focused on future-ready solutions, James Fleischmann is approaching $50 million. To achieve remarkable growth and impact, he combines innovation, scalability, and AI efficiency.
Felicitaciones por el ARR 💪 ¿Usaste ads, comunidad o SEO para escalar? Me interesa porque estoy lanzando un marketplace y quiero aprender de los que ya pasaron por ese proceso.
The focus on balancing rapid innovation with a solid technical infrastructure is truly remarkable. I’m curious, how do you internally manage the trade-off between the need to release immediate features and the commitment to building modular foundations for the future? Do you have any criteria or framework that helps you make these strategic decisions in such a dynamic environment?
Absolutely loved this. The mix of deep technical foundation with sharp product intuition really comes through — especially the part about resisting UI polish early on and focusing on scalable infra first. That kind of discipline is rare.
Curious: how do you balance shipping features for current use cases while still laying down foundations for “the next wave”? Do you have a framework internally for deciding what’s tactical vs. what’s foundational?
This is an incredible journey proof that understanding tech deeply and communicating it clearly is what separates great tools from great businesses.
I work with growing AI and dev focused companies to turn complex processes into compelling product stories, technical landing pages, and emails that don’t just explain they convert.
If Augment or any team reading this wants words that ship alongside the code, I’d love to collaborate.
Insane growth!
How did you handle the first wave of user feedback? I always wonder how AI products manage expectations when the tech moves so fast.
it's refreshing to see real distribution strategy.
i liked the design of the picture
Love seeing your growth story here. Curious: how did you drive your first 100 users?
Really love this perspective, especially the part about building for the next wave while shipping for the current one. AI definitely feels like it's evolving on fast-forward, and teams that aren’t thinking modular or future-proof risk getting left behind. Totally agree that balancing rapid execution with technical foundations is what separates the hype from real product-market fit. This advice is applicable across different hyper competitive industries where building for the future will help create a moat of sorts
Really inspiring story and a solid reminder that building infrastructure before the hype wave hits is what separates a lucky launch from a durable business. The focus on deep codebase understanding and developer-first design really shows a clear grasp of the pain points in real engineering environments. Also appreciate the honesty about the tempo and mindset needed to survive in AI — it’s not for everyone, but the ones who thrive in it will build the future.
Wow your story is so inspiring...l just new in here n dont know wht should do,but l hope l can be like u
Loved the part about optimizing for distribution from day one. Many founders overbuild before validating distribution, and it’s refreshing to see someone at this scale still focused on shipping fast and testing early.
Hey Igor, thanks for the article. How do you personally stay ahead of the curve in such a fast-moving field like AI? Any habits or filters you use to separate signal from noise?
Great insights on building for next year, Igor! Your "bet forward" mentality really resonates.
I'm working on Avery (AI task management via WhatsApp) and face similar challenges around rapid AI evolution. Like you, I've learned that the hard systems work has to come early. For us, that meant building proactive follow-up logic and contextual memory before polishing the chat interface.
Your point about "modular retrieval layers and model abstraction seams" hits home. We're constantly balancing shipping for today's capabilities while architecting for whatever breakthrough drops next quarter. The half-life measured in quarters is real.
Question: How do you handle the tension between deep codebase understanding (your strength) and the pressure to ship AI features that feel magical to users? I find that explaining the "why" behind intelligent behavior is almost as hard as building it.
The "try first" approach is smart too. We're seeing similar patterns where people need to experience the AI difference firsthand, especially when the market is so noisy with AI hype.
Congrats on the growth. Building something this technical while keeping that rapid iteration speed is no joke.
Incredible story Igor! Your "bet forward" approach resonates deeply.
I'm building AlbumForge (photo software with 1=1 giving model) while maintaining my therapy practice. Your point about building for next year really hits home. We just went through similar challenges scaling to 44 languages while maintaining quality - the temptation is always to optimize for today's constraints instead of tomorrow's possibilities.
Your Context Engine handling hundreds of thousands of files is impressive. We found that the hardest part isn't the technical scale, but maintaining the "why" behind what you're building when everything moves so fast.
Question: As you approach $50M, how do you keep that "deep focus" culture you mentioned? I'm curious about balancing rapid iteration with intentional product decisions.
Thanks for sharing this - proof that understanding technology + customer + market deeply before choosing direction really works!
Using an AI coding agent, James Fleischmann discusses achieving $50M through long-term value, team alignment, and building tools that meet the needs of future developers. The key to success is strategic foresight.