Six months in, Gravity is still pre-launch, and that is the honest headline. There are no customers, no revenue, and no growth chart to show. What there is instead is a set of decisions made deliberately and a product being hardened against a quality bar before we open the doors. This is a build-in-public retrospective from the Gravity team, founded by Aryan Agarwal at XAI Technologies in Bangalore, on the first half of 2026: what we chose, why we chose it, and what we would do differently.
We write these because the reasoning is more useful than the outcome at this stage. Anyone can publish a victory lap after launch. Far fewer share the calls they made while the result was still uncertain, which is exactly when the thinking is worth recording. If you have read our earlier notes on three startups and three shutdowns or on bootstrapping an AI agent platform, this post sits on top of both: the wider arc of the company so far, told through the decisions that defined it.
Where we are at six months
Gravity is an AI agent platform. A user describes an outcome in plain words, an expert-built agent runs it, and the finished result comes back in about 60 seconds. The tagline, "AI Agents in 60 seconds," is a promise we are still earning rather than a feature we are advertising. As of mid-2026 we are on a waitlist, with agents in testing and a content engine running well ahead of the product.
The market context made the timing feel right. Adoption of generative AI inside companies moved from experiment to line-of-business use over the past two years, with surveys from McKinsey's State of AI reporting that a majority of organizations now use the technology in at least one function. The gap we kept seeing was the distance between a capable model and a finished outcome a non-technical person can actually trust, a translation problem that policy and research bodies like the OECD AI Policy Observatory frame as the hard part of turning AI capability into real-world value. Closing that gap, not adding another chat box, is the reason Gravity exists. The rest of this review is the set of decisions that followed from taking that gap seriously.
The decision to be a platform
The first and most consequential decision was what kind of company to be. It would have been easy to frame Gravity as a marketplace: a place where independent builders publish agents and we take a cut. We chose deliberately against that. Gravity is a platform where the company is responsible for the service. Users prompt and run expert-built agents. Builders build and maintain those agents for Gravity as service providers. Gravity runs the agents, carries the cost of running them, and stands behind the output.
The difference is not cosmetic. In a marketplace, accountability is diffuse: if an agent gives a bad answer, the platform shrugs and points at the seller. We did not want a user to ever have to wonder who is responsible, so we put that responsibility on Gravity by design. That single choice cascaded into everything downstream, from how agents get tested to how we talk about the product. It is also why we describe the quality bar in terms of what Gravity guarantees rather than what a builder promises.
This was not the framing we started the year with, and changing it mid-stream was uncomfortable. Aligning the public positioning to the platform model took a deliberate revision, and we kept the older marketplace language out of the site once we made the call. The lesson we took is that positioning is a product decision, not a marketing one. Get it wrong and every later choice inherits the confusion. Get it right and the rest of the roadmap tends to fall into place.
Pay-per-use over subscriptions
The second big decision was pricing. The default for software is a monthly subscription, and we considered it seriously before rejecting it. The value of an agent is the run: the moment it reads your data, does the work, and hands back a result. A subscription charges for access regardless of whether that moment ever happens, which punishes the occasional user and rewards us for shelf-ware. That felt backwards for a product whose whole pitch is "describe a task and it gets done."
So we chose pay-per-use credits: one dollar buys 1,000 credits, and you only pay when an agent actually runs. Cost tracks value instead of a calendar. We wrote up the full reasoning in why we chose credits over subscriptions, and we have started stress-testing the model itself in the credit pricing model six weeks in. The honest caveat: usage-based pricing is harder to forecast than a subscription, both for us and for the customer, and we will be watching closely how it behaves once real workloads run. We think the alignment is worth the added uncertainty. We will know for sure only after launch.
The 60-second promise and the quality bar
"AI Agents in 60 seconds" is a promise about experience: describe an outcome, get a finished result, no setup or dashboard to operate. A promise like that is only as good as the agent behind it, which is why the most demanding work of the past six months has gone into the quality bar rather than into features. An agent that returns a fast wrong answer is worse than no agent at all, because it spends the user's trust on the first run.
Every agent capability is tested extensively before it ships, against the kinds of inputs real users will throw at it, including the messy and the adversarial. We described the standard in the Gravity agent quality bar explained, and the testing discipline in how we test AI agents. The principle is simple to state and slow to satisfy: an agent does not ship until it clears the bar, and the bar is set by what Gravity is willing to be accountable for, not by a launch deadline. Reliability research generally finds that the hard part of AI systems is consistent behavior on the long tail of edge cases, a pattern echoed in the Stanford HAI AI Index. That long tail is where most of our testing time goes.
The pre-launch content engine
The decision that surprises people most is that we started publishing heavily before we had a product to sell. The reasoning: search visibility and topical authority compound slowly, so the cost of starting late is permanent. Building the content engine during pre-launch means the library is already maturing when the product opens, rather than starting from zero on launch day.
It has not been a clean win, and we have said so. We wrote candidly about the early disappointment in lessons from 500 blog posts and no traffic, because pretending content compounds overnight would be dishonest and would set a bad expectation for anyone reading along. The qualitative read at six months: the engine is producing a deep, internally linked library on AI agents, indexing is improving, and the early signs of compounding are showing up, slowly, in the way they tend to. We are not going to dress thin early numbers as a breakout. The bet is a long one, and we are still early in it.
What we got wrong, and the road to launch
A few honest corrections. We underestimated how much the positioning change would ripple through everything we had already written and built, and unwinding the marketplace framing took longer than expected. We were slower than we would have liked to accept that content is a multi-quarter investment, not a multi-week one. And we have learned, repeatedly, that the quality bar costs more time than any plan accounts for, because the last ten percent of reliability is most of the work.
What is left before launch is straightforward to name and hard to finish: get the priority agents fully through the quality bar, harden the pay-per-use system against real usage, and bring the waitlist into a controlled early access rather than a single big-bang open. We keep a running account of that work in the launch readiness update, which we will keep updating as the picture changes. Six months in, the company is exactly what it claims to be: pre-launch, deliberate about its decisions, and willing to show the reasoning before the results exist.
How Gravity handles the work behind a review like this
A retrospective like this involves a lot of unglamorous synthesis: pulling notes together, drafting, fact-checking claims, and tightening copy. That is precisely the kind of outcome Gravity is built to deliver once it launches. Gravity is an AI agent platform. You describe the outcome in plain words, say, summarize these six months of build notes into an honest review, and an expert-built agent runs it and hands back the finished draft in about 60 seconds, instead of a blank document you have to fill yourself.
The model is pay per use: one dollar equals 1,000 credits, and you only pay when the agent actually runs. There is no seat to buy and no dashboard to learn. Gravity runs and maintains the agent, carries the cost of running it, and is responsible for the output, which is the same accountability principle this whole review is built around. If you are new to the idea, setting up your first AI agent walks through going from a plain-language description to a running workflow, and what is an AI agent explains why this counts as agentic work rather than a one-off prompt. The product is on a waitlist for now; this post is part of how we show the thinking while we finish the build.
FAQ
Is Gravity launched yet?
Not yet. As of mid-2026 Gravity is pre-launch and running a waitlist. We are still hardening agents against the quality bar before we open the doors. This review is a build-in-public look at the first six months of decisions, not a results report, because there are no live customers or revenue yet.
Why does Gravity call itself a platform and not a marketplace?
Because Gravity is responsible for the service. Users prompt and run expert-built agents. Builders build and maintain agents for Gravity as service providers. Gravity runs the agents, carries the cost, and stands behind the output. That accountability is the whole point, and a marketplace framing would have put it in the wrong place.
Why did Gravity choose pay-per-use credits instead of a subscription?
Because the value of an agent is the run, not the seat. Subscriptions charge for access whether or not you use it, which punishes occasional use. Pay-per-use credits, one dollar for 1,000 credits, mean you only pay when an agent actually does work, so cost tracks value rather than a calendar.
What is the 60-second promise?
It is the experience we are building toward: you describe an outcome in plain words and an expert-built agent hands back the finished result in about 60 seconds, with no setup or dashboard to operate. The promise sets the bar for both product design and the quality testing every agent has to clear before it ships.
How can I follow Gravity's progress to launch?
Join the waitlist to get launch notice, and read the blog, where we publish build-in-public updates on positioning, pricing, the quality bar, and launch readiness. We share the reasoning behind decisions as we make them rather than waiting for a polished post-launch story, so the blog is the most current record.