Imbue wants you in the driver's seat.

It is a Tuesday in San Francisco, and somewhere inside Imbue a software engineer is doing something quietly radical: trusting a machine to write code, and then watching it the way you'd watch a new hire - closely, and with a hand near the undo button.

That image is the whole company in miniature. Imbue is an AI research lab, currently around 20 to 78 people depending on who's counting, building agents that can reason and write software. It is led by CEO Kanjun Qiu and CTO Josh Albrecht. It is worth more than a billion dollars. And its entire pitch rests on a sentiment that sounds almost quaint in 2026: that the person using the tool should remain its boss.

Most AI companies sell you the dream of handing things off. Imbue sells you the dream of staying involved - which is either heresy or common sense, depending on how your last week of debugging went.

The problem they sawThe handoff that never quite works

Here is the tension Imbue has decided to live inside. AI agents are getting good enough to do real work - and the moment they do, the temptation is to stop looking. To let the agent run, merge the diff, ship it, and hope. The industry calls this autonomy. Engineers who've cleaned up after an over-confident agent call it something less flattering.

Imbue's read is that the dangerous part of agentic coding isn't the code the agent writes. It's the code you don't read. An agent that confidently introduces a subtle bug across forty files is not a productivity gain; it's a liability with good manners.

The founders' betTwo people, one inconvenient conviction

Kanjun Qiu founded the company in 2021 - then called Generally Intelligent - with co-founder Josh Albrecht. Their bet was unfashionable: instead of racing toward fully autonomous systems, build agents you can understand, modify, and verify, so they work only in your interest. Qiu has described the goal as rekindling the dream of the personal computer, where the machine amplified the individual rather than the platform.

In September 2023 the bet got a number attached to it. Imbue raised a $200 million Series B led by the Astera Institute, with Nvidia, Cruise CEO Kyle Vogt, and Notion co-founder Simon Last joining in. The round valued the company north of a billion dollars and came with a new name. Generally Intelligent became Imbue - a verb, pointedly, about putting human values into machines.

The productSculptor, or: agents with a leash

If the mission is abstract, Sculptor is the part you can download. It's a UI for running coding agents in parallel - and the design choice that makes it interesting is also the least glamorous one: containers.

Every agent in Sculptor runs in its own isolated container. They can all execute code at the same time without trampling each other or your laptop. Unlike git worktrees, which share your local environment and force you to reinstall dependencies for each agent, Sculptor keeps your machine clean and lets you slide between agents without losing the thread. When an agent's work looks good, you bring its changes straight into your IDE to test and review.

Translation for the skeptics: the whole point is that you can fire any agent that misbehaves, instantly, without it having touched anything real.

The proofNumbers, backers, and the hard part

Imbue's evidence is partly financial and partly architectural. The financial part is easy to state: a $1B+ valuation reached with a team that, at the time of the raise, was reported to be around twenty people. That is an unusually small denominator for an unusually large number.

The architectural part is where Imbue is honest about being early. Sculptor is in beta, built for teams willing to deploy fast and report back. Backers like Nvidia bring compute as much as cash - the H100 cluster is the kind of infrastructure that lets a small team iterate on training data, architecture and reasoning mechanisms quickly. The customers, for now, are engineers and engineering teams adventurous enough to put coding agents to real work.

The missionA billion-dollar argument about control

Strip away the funding and the GPUs and Imbue is making a single argument: that technology should be beholden to the people who use it, with no ulterior motives. It's the sort of line that's easy to put on a careers page and hard to keep when incentives pull the other way. Imbue's answer is to build the conviction into the product - agents in containers you control, models you can inspect, tools released into the open.

Whether that's principled or simply a sharp wedge against closed platforms is a fair question. The honest answer is probably: both. The best missions usually are.

Why it matters tomorrowBack to that Tuesday

Return to the engineer at the start - the one watching the machine write code with a hand near the undo button. Imbue's whole reason for existing is to make that posture obsolete in the good way: not by removing the human, but by making the machine trustworthy enough that watching it feels less like supervision and more like collaboration.

That's the bet. Agents that reason instead of guess. Tools that keep you in the loop instead of cutting you out. A future where, when the AI writes your software, you're still the one who decides what ships. It is a smaller, stranger ambition than "replace the engineer." It might also be the more durable one.

WatchInterviews & demos

Go deeperFind Imbue