When we founded Imbue in 2021, we set out to build an AGI future where humans remain at the helm, shaping powerful AI systems rather than being subordinated to them.
We saw AI’s potential to create abundance and achieve what’s currently impossible. But we also noticed many people’s deep worry that as AI grew more capable, it might marginalize humans — eroding our power, agency, and freedom to shape our lives.
We’ve come to understand something crucial: the core challenge in protecting human agency comes down to managing how AI shifts power.
AI makes software dramatically more powerful. By default, this power naturally flows to those who build and own those systems. Left unchecked, this concentration of power leads to exploitation and disempowerment. We’ve already seen this pattern: tech platforms use basic AI to hijack our attention, creating addictive experiences we at once crave and resent. As a result, we’ve lost control of our digital lives and have almost no say in how these systems operate or use our data.
We initially believed smaller AI agents that helped us accomplish tasks would naturally distribute power. But we realized even these agents could undermine human agency: like tech platforms, we still controlled algorithms that made decisions about people’s lives. This insight changed our approach: real empowerment doesn’t come from centrally-controlled agents, but from enabling people to build and shape their own AI software.
Imbue’s mission is to empower humans in an age of AI by creating powerful computing tools for individuals. Like the right to vote in a democracy, we believe the ability to create, modify, and control software and agents gives us a voice in this era of powerful AI — so that humans can be actors, not acted upon. We envision a future where AI serves as a genuine tool, empowering us to pursue our own goals, rather than subjecting us to misaligned incentives and systems beyond our control.
This requires a shift in the philosophy of building AI — not AI agents to replace people, but to help people build. In an era racing to make machines that replace humans, we want to reveal the builder in every human.
Today, we see glimpses of this possibility: AI coding tools seem tantalizingly close to letting anyone build software simply by describing it. Unfortunately, what starts as a spark of excitement often turns to frustration as the codebase grows unwieldy, adding a simple feature requires rethinking the entire data model, and hours are wasted chasing down bugs in code we didn’t write.
This is fine for toy apps and quick demos, but falls short when it comes to creating genuinely powerful software we can rely on to serve our real purposes. Without stronger creation tools, we’ll remain dependent on other entities whose incentives may not match our own.
To achieve this, code generation isn’t enough; the hard part is debugging, maintenance, and managing complexity. As a field of software engineering, we’ve developed best practices for architecting reliable systems, testing, and managing changes — but today’s LLM workflows rarely incorporate these practices.
We’re creating a better way to build software with AI — one that embeds engineering best practices directly into AI-assisted software development, so more people can easily create robust, dependable software for their lives.
Our initial product is a coding agent environment that helps engineers write healthier code faster with LLMs by making it easy to write tests, identify and fix issues, and execute LLM-generated code safely.
You can see more details and try it out here.
Our longer-term research focuses on methods for building trust in AI-generated code, like automatic code verification, catching and fixing bad code, and identifying robust libraries to use by default. Once our systems are better at writing, verifying, and fixing code, people can focus more on what they want to create rather than the code itself — making software creation accessible to many more people.
Ultimately, we want to enable the engineering community to embed our collective knowledge of software craft into an open environment of coding agents that upholds our shared principles. Engineering is at heart a social practice, and we want tools that help people produce robust, reliable, understandable code. Those qualities of code are human-centric, not machine-centric.
Imbuing best practices makes it much easier for anyone to build software that lasts, opening the door for many more people to participate in the software economy. Instead of waiting for companies to build for us, we’ll be able to make our own idiosyncratic tools for ourselves and our communities — for example, I’d love an app that protects my grandmother from scam calls in Chinese (something surprisingly hard to find today).
When we can create and control software and algorithms, power shifts. At the most basic level, being able to build our own interface to services lets us resist algorithms and platforms we currently cannot opt out of. Platforms have to serve our interests to keep us engaged; otherwise, we can just pivot to our own solutions. The same dynamic applies when faced with other people’s AI agents — we can have our own agents that safeguard our attention and time against theirs.
But technology isn’t the whole answer. We also need the right laws and societal structures — ones that let our agents be as powerful as those controlled by companies with lots of data (for example, by letting us get our own data out of corporate silos), and that protect us from other entities’ agents when they impinge on our freedoms (for example, by trying to capture our attention or subtly shape our decision to buy something). This is the core of our policy work at Imbue: to safeguard individual rights in an automated world and uphold democratic principles against power concentration.
Technology’s highest purpose has never been to replace human capability, but to amplify what was already there. We believe creativity lies within every person, waiting to be unlocked. The world’s most meaningful software is still trapped in the human imagination, locked behind the barrier between what billions of people can imagine, and what they can actually create.
And if we free it, we can create something better than our current trajectory: a world more democratic, more open, more free. A world where we imbue our will into machines to shape our lives and institutions, where we collectively direct our agents toward solving what matters most to us.
Instead of being replaced by AI, we can use it to nurture what is best within each and every one of us: the capacity for creation, for meaning, for love, for joy, for beauty, for awe.
This is the human future we can fight for. This is what AI is for.