Here is a thing that is true about almost every software company: at some point, an engineer sits down to write a small, honest piece of code - send this email, resize this image, charge this card - and then spends the next three weeks building the machinery around it. A queue. A retry loop. A cron job. A dead-letter dashboard nobody looks at until it is 3am and everything is on fire. The original code was five lines. The scaffolding was five hundred.
Inngest exists because two engineers decided this was a bad trade and, unusually, did something about it. Tony Holdstock-Brown, formerly of Docker, and Dan Farrelly, formerly the CTO of Buffer, founded the company in San Francisco in 2021 with a proposition that sounds almost too tidy: what if reliability were just a property of your code, the way a type is, rather than a separate system you have to operate?
What It Actually Does
The technical heart of Inngest is durable execution. You write a function. You break it into steps. Each step is a small, transactional unit that Inngest remembers. If step three fails, Inngest retries step three - it does not re-run steps one and two, does not re-charge the card, does not re-send the email. State is persisted. A function can go to sleep for a week and wake up exactly where it left off, which is the kind of thing that sounds trivial until you have tried to build it yourself.
Around that core sits what the company calls flow control: concurrency limits, per-tenant fairness so one noisy customer cannot starve the rest, rate limiting, throttling, debouncing, batching, and prioritization. These are the unglamorous levers that separate a demo from a system that survives contact with real traffic. Inngest turns each of them into a line of configuration rather than a weekend of plumbing.
The delivery mechanism is deliberately contrarian. Inngest invokes your code over HTTP. There is no worker fleet to run, no long-lived pool of processes to babysit - a pointed contrast with Temporal, the heavyweight incumbent in this space. Your functions live wherever you already deploy: Vercel, AWS, Cloudflare, Netlify, a laptop. The core engine is open source, so you can run the entire thing locally and watch it work.
The AI Turn
Then the ground shifted, as it does. AI agents turned out to be, from an infrastructure standpoint, exactly the kind of long-running, failure-prone, wait-for-a-human, call-six-tools workload that durable execution was built for. An agent that times out halfway through a task, or needs to pause for a day while a person approves something, is precisely the problem Inngest already solved for ordinary backends.
So in 2025 the company shipped AgentKit, a TypeScript framework for building networks of agents with deterministic routing, shared state, and support for OpenAI, Anthropic, and Gemini models. The positioning writes itself: the reliability layer for the AI era. Investors, who had watched the company grow from a $3M seed into a platform running over 100 million executions a day, agreed - to the tune of a $21M Series A led by Altimeter, with a16z and Vercel's Guillermo Rauch returning.
What is quietly remarkable about Inngest is how little it wants you to notice it. The best outcome, for its users, is that a whole category of work - queues, retries, recovery - simply stops being a thing they think about. SoundCloud, Tripadvisor, Resend, Contentful, GitBook, and Replit have taken the deal. The product's ambition is measured not in features added but in code deleted.