Talent density. Amazon hired the cofounding team for AGI work under Rohit Prasad (TechCrunch). Licensing without owning the entity. Amazon got models and agent tech while Adept's corporate shell continued (Adept).

Adept: $400M+ agent startup whose founders left for Amazon

Q: What didn't work for Adept, and what are the risks?

Did not bring a general product to market despite extensive testing, per TechCrunch (TechCrunch). Training costs. Semafor: skyrocketing model-training costs burdened the original AGI plan (Semafor). Investor outcome was return of capital, not venture-scale upside (Semafor). Funding and "made whole" details rely on Adept's blog plus TechCrunch/Semafor reporting with anonymous sources for payment terms.

Key takeaways

Founded ~2022 to train multimodal models into agents that operate software; raised over $415M at about a $1B valuation.
28 Jun 2024: cofounders and much of the team join Amazon AGI; Amazon licenses Adept models and agent tech; Zach Brock becomes CEO of the remaining company.
Semafor later reported investors who put ~$414M in were roughly made whole via Amazon-related payments — a talent deal, not a classic acquisition.

What it is

Adept set out to build AI agents that use existing computer software via natural language — an "AI teammate" for workplace tools (TechCrunch, Adept).

Timeline

~2022: Adept founded by researchers including former OpenAI VP David Luan; mission to train multimodal foundation models, fine-tune them into agents, and ship enterprise products (Adept).
Through 2024: raises over $415M from backers including Nvidia, Atlassian, Workday and Greylock; valuation around $1B. TechCrunch notes early cofounder departures and little shipped product despite long betas (TechCrunch).
28 Jun 2024: Adept announces cofounders and some of the team are joining Amazon's AGI organization; Amazon licenses Adept's agent technology, multimodal models and some datasets; Zach Brock becomes CEO as Adept refocuses on agentic solutions (Adept, TechCrunch).
2 Aug 2024: Semafor reports investors (who put about $414M in) will roughly recoup their investment via arrangements tied to Amazon's payments, while Adept keeps ~one-third of employees and about $25M (Semafor).

Why it worked (for Amazon, not as an independent product)

Talent density. Amazon hired the cofounding team for AGI work under Rohit Prasad (TechCrunch).
Licensing without owning the entity. Amazon got models and agent tech while Adept's corporate shell continued (Adept).

What to learn

Foundation-model ambitions plus an enterprise product can force an impossible fundraising dual track — Adept said continuing both would consume attention better spent on agents (Adept).
"Acqui-hire via license" can return capital to investors without a formal acquisition; regulators may still scrutinize it (Semafor).
Shipping matters: raising hundreds of millions without a clear product market makes a soft landing more likely than independence (TechCrunch).

Risks and what didn't work

Did not bring a general product to market despite extensive testing, per TechCrunch (TechCrunch).
Training costs. Semafor: skyrocketing model-training costs burdened the original AGI plan (Semafor).
Investor outcome was return of capital, not venture-scale upside (Semafor).
Funding and "made whole" details rely on Adept's blog plus TechCrunch/Semafor reporting with anonymous sources for payment terms.

FAQ

Short answers drawn from the sections above.

What is Adept?

Adept set out to build AI agents that use existing computer software via natural language — an "AI teammate" for workplace tools (TechCrunch, Adept).

Why did Adept work?

Talent density. Amazon hired the cofounding team for AGI work under Rohit Prasad (TechCrunch).
Licensing without owning the entity. Amazon got models and agent tech while Adept's corporate shell continued (Adept).

What can indie builders learn from Adept?

Foundation-model ambitions plus an enterprise product can force an impossible fundraising dual track — Adept said continuing both would consume attention better spent on agents (Adept).
"Acqui-hire via license" can return capital to investors without a formal acquisition; regulators may still scrutinize it (Semafor).
Shipping matters: raising hundreds of millions without a clear product market makes a soft landing more likely than independence (TechCrunch).

What didn't work for Adept, and what are the risks?

Did not bring a general product to market despite extensive testing, per TechCrunch (TechCrunch).
Training costs. Semafor: skyrocketing model-training costs burdened the original AGI plan (Semafor).
Investor outcome was return of capital, not venture-scale upside (Semafor).
Funding and "made whole" details rely on Adept's blog plus TechCrunch/Semafor reporting with anonymous sources for payment terms.

Sources

Every link cited above, in order of first use.

1.TechCrunch (techcrunch.com)
2.Adept (adept.ai)
3.Semafor (semafor.com)