Adept, which aims to build AI systems that can understand and automate any software process, emerges from stealth with $65M led by Greylock and Addition
Context & Ripple Effects
Adept's stealth exit is the opening move in a bet that the next interface for software is a model that operates it: rather than building another app, the company wants an AI that can understand and drive any software process on a user's behalf. Backers Greylock and Addition are underwriting that thesis at the seed-to-A stage, before the category has a name.
The bet aged quickly into a pattern. Within a year Adept had converted this round into a $350M Series B at a $1B+ post-money valuation, and the agent stack it implied began filling in around it — most visibly with Adapter's later emergence selling the data-control infrastructure those agents need. Parallel rounds like Augment's $227M Series B show the same capital thesis applied to coding specifically.
First-order effects
- Greylock and Addition secure early positions in a company whose product category — AI executing arbitrary software workflows — had no established leader at the time of the round.
- Adept gains the runway to train and evaluate models against real software environments, the expensive part of the problem that separates it from thin workflow-automation wrappers.
Second-order effects
- The rapid follow-on at a $1B+ valuation signals to other investors that agent-style startups can raise at platform multiples, pulling new entrants like Adapter into adjacent layers of the same stack.
- Software vendors whose products would be driven by such agents face a choice between treating them as a threat to their UIs or exposing APIs that make their tools agent-controllable first.
Third-order effects
- If models that operate any software process mature, the work of gluing enterprise systems together shifts from human integrators and outsourced services toward model vendors — consistent with reported pressure on formulaic IT-services work and entry-level professional-services roles.
- A working universal software operator would also concentrate value in whoever controls the agent's access to data and permissions, which is why infrastructure plays like Adapter's are emerging as a distinct layer rather than a feature.
The trend: Venture capital is funding a new class of AI-native companies whose product is not an app but an operator of all apps, with round sizes escalating as the agent-stack layers specialize.