This is an unusual comparison, because by 2026 Adept is not really a product you can choose. Adept was one of the most admired startups in the agent space: its ACT-1 demos, where an AI operated real software the way a person would, did as much as anything to define what people now mean by an AI agent. Then in 2024 Amazon hired its founders and licensed its technology in a reverse acqui-hire, and Adept continued as a smaller team focused on enterprise solutions rather than the consumer-facing action agent it first showed (TechCrunch, 2025).

I am not going to dunk on Adept. I have shut down three of my own startups, so I read this story with more empathy than schadenfreude. But it is genuinely instructive, and it says something real about why I am building Gravity as a marketplace rather than a single research-first lab. So this piece is half comparison and half lesson.

What Adept was, and what happened

Adept set out to build a foundational action model: an AI trained not just to generate text but to operate software, click, type, and navigate tools the way a knowledge worker does. The ambition was huge and the early demos were electric. The problem was the one that has humbled many frontier labs: training models at that scale is brutally expensive, and a research-first company has to fund that burn while also becoming a product company (eesel AI, retrieved 2026).

The Amazon reverse acqui-hire

In 2024, rather than continue that race alone, Adept's leadership joined Amazon. The structure was a reverse acqui-hire: Amazon hired the founders and key staff and licensed Adept's technology rather than buying the company, and reporting indicated investors were roughly made whole while Adept itself received around twenty-five million dollars and continued with a smaller team (Semafor, 2024). Co-founder David Luan went on to lead an Amazon AGI lab, which later shipped Nova Act, a model and toolkit for browser-acting agents.

Where things stand in 2026

By 2026 the picture had shifted again, with reporting that several of the Adept co-founders who joined Amazon had since departed (GeekWire, 2026). The throughline is simple and a little sad: a category-defining vision, a team good enough that Amazon wanted it, and a product that nonetheless did not survive in the form it was first sold. That is not failure of talent. It is how hard the model-building path is.

What Gravity does differently

Gravity is not trying to build a foundational action model. That is the opposite of my bet. Gravity is a marketplace that sits on top of existing models and distributes finished, tested agents built by many different experts. You describe an outcome in plain language and run an agent an expert already built, in about 60 seconds, paying per use, where one dollar buys 1,000 credits. See how Gravity works.

The distinction is between inventing the engine and assembling a lot of cars people can drive today. Adept tried to invent the engine, which is capital-intensive, winner-take-most, and existential if the funding math turns. Gravity assembles a catalogue on engines that already exist, so the hard, expensive model research is not a bet the whole company lives or dies on. If you want the conceptual line between a model and an agent, see AI agent vs LLM distinction.

The three-sided marketplace

Gravity has three sides. Users run agents and pay per run. Builders publish agents and earn 20% of every run as pure profit. Creators earn 10% on runs from people they refer. The structural point relative to Adept: value and risk are distributed across many builders, not concentrated in one lab's research roadmap. For the builder economics, see how to monetize AI agents.

Side-by-side comparison

Because Adept is no longer a buyable consumer product, this table compares the models and the bets, not feature checklists.

DimensionAdept AIGravity
Core betBuild a foundational action modelDistribute finished agents on existing models
Layer in the stackThe model itselfThe agent and the marketplace
Capital intensityVery high (frontier training)Low relative to model training
Status (2026)Reverse-acqui-hired by Amazon; smaller, enterprise focusPre-launch, actively building
Who builds the agentAdept's research teamMany vetted expert builders
Availability to a buyerNot a consumer product to sign up forWaitlist now, pay-per-use at launch
Concentration of riskOne lab, one roadmapSpread across many builders
Earning modelVenture-funded researchBuilders earn 20%, creators 10%

The contrast is not which product has more features. It is which structure survives contact with the brutal economics of this field, and which leaves a buyer holding the bag if a single company's bet does not pay off.

Build the model vs distribute the agent

Adept and Gravity sit at different layers, and the layer you operate at shapes your risk. Building a frontier model is the most capital-hungry, highest-variance bet in the whole stack; even brilliant teams can run out of runway before the product catches up. Distributing finished agents on top of models that other well-funded labs are already racing to improve is a fundamentally lighter bet, and it gets better automatically as the underlying models improve.

This is the same build-versus-buy logic I apply to customers in build vs buy AI agent, turned on the company itself. Adept chose to build the hardest layer. Gravity chooses to assemble and distribute on top, which is less heroic and, I would argue, more durable for the thing a user actually depends on: the work continuing to run tomorrow.

The longevity lesson for buyers

Here is the practical takeaway, and it applies far beyond Adept. When you standardize a workflow on a vendor, you are also betting on that vendor still being here, and still pointed the same way, in a year. A single research-first lab is a concentrated bet: if it gets acquired, pivots, or runs out of money, the workflow you built on it can evaporate. We watched exactly that happen to a company good enough that Amazon wanted its whole team.

A marketplace is a different risk profile. No single builder's fate decides whether the catalogue keeps working, and if one agent's maker moves on, another can fill the gap. That does not make a marketplace immortal, and I would never claim Gravity is risk-free as a pre-launch company. But spreading the building across many experts is a deliberate hedge against the single-point-of-failure problem the Adept story illustrates. For where I think this all heads, see AI agent future trends 2026.

If you wanted what Adept promised

Plenty of people fell for Adept's promise for good reason: AI that just does the task on your screen is the dream. If that is what drew you, the honest path in 2026 is not to wait for Adept to re-emerge in its original form. It is to look at products that deliver finished outcomes today. Some of that lives inside Amazon's Nova Act lineage; some lives in finished agents you can run now. A marketplace of expert-built task agents is one practical way to get the outcome without betting on a single lab's survival.

When Gravity fits

Three signals say Gravity is worth your attention. First, you care about the outcome, not about owning a model; you want the task done. Second, you value not betting your workflow on one company's research runway. Third, you would rather pay per run for a tested agent than fund a frontier bet through a subscription. For why I made this bet, including the three startups I shut down before it, see about Gravity and my Super AI postmortem. For a closer comparison on demo-grade versus operations-grade autonomy, see Gravity vs Manus.

Frequently asked questions

Can I still use Adept AI in 2026?

Not in the way the early ACT-1 demos suggested. In 2024 Amazon hired Adept's founders and licensed its technology in a reverse acqui-hire, and Adept continued as a smaller team focused on enterprise solutions rather than the consumer-facing action agent it first showed. If you came looking for a product to sign up for, verify Adept's current offering directly before planning around it.

What happened to Adept AI?

Adept pioneered action models, AI that operates software like a person, and its ACT-1 demos helped define the agent category. Facing the high cost of training frontier models, in 2024 Amazon hired its co-founders, who went on to lead an Amazon AGI lab, and Adept pivoted to a smaller enterprise-solutions focus. The vision was influential; sustaining it as an independent product proved hard.

What is the difference between Adept and Gravity?

Adept was a research-first company building a foundational action model. Gravity is a marketplace that distributes finished, tested agents built by many experts on top of existing models. Adept aimed to invent the engine; Gravity assembles a catalogue of cars you can drive today. The two sit at very different layers of the agent stack.

Is Gravity an Adept alternative?

If you were drawn to Adept's promise of AI that gets real tasks done for you, then a marketplace of expert-built task agents is a practical alternative, because it delivers finished outcomes today rather than a research model you would have to build on. It is not a like-for-like swap, since Adept built models and Gravity distributes agents, but it serves the same underlying want.

What does the Adept story teach buyers?

It is a reminder that platform longevity is part of any purchase. Betting your workflow on a single research-first lab carries real risk if that lab gets acquired or pivots. A marketplace spreads that risk across many builders and finished agents, so no single company's fate decides whether the work you depend on keeps running.

Who acquired Adept AI?

Amazon did not buy Adept outright. In 2024 it hired Adept's founders and key staff and licensed the company's technology, a structure often called a reverse acqui-hire. Adept's co-founder David Luan went on to lead an Amazon AGI lab. Reporting noted that Adept itself continued with a smaller team and a narrower enterprise focus.

Three takeaways before you close this tab

Sources