Agents are getting better at working alone, and the improvement is measurable.
METR have been tracking how long a task an agent can finish on its own, in units of human working time. Since the launch of ChatGPT, task length has been doubling roughly every six months:
Frontier models are now somewhere around a full working day of unattended effort.
Maybe the curve eventually plateaus. We don't think that's the right assumption to build around, so we're building for a world where autonomous agents continue to become more capable.
It's also worth remembering what this benchmark measures. It only includes tasks that a sufficiently capable agent can complete end to end, without needing to stop for permission or wait for input from a human.
We’re making sure that setting up observability from scratch is one of those tasks.
One POST, one org, 24 hours
There's now an unauthenticated endpoint that lets an agent create a real Axiom organization. No signup, no email confirmation, no messing about:
This endpoint returns an org identifier, an API token with full permissions, an expiry timestamp, and a claim URL:
The agent can then create datasets, ingest events, query, build dashboards, and set up monitoring. It's a real org, and behaves like any other.
If a human follows the claim URL and signs in within 24 hours of creation, the org becomes permanent and ownership transfers to the signed-in user. If nobody does, we delete it.
Two things made us comfortable opening this door. First, Axiom is purpose-built for efficiency, from a custom ingest pipeline to storing your data in an industry-leading, highly efficient storage format on object storage, all the way to serverless queries that don't need upfront provisioning. Second, we've run a genuinely useful free tier for years, with over 100K users, so free-tier abuse is not a hypothetical we're gaming out for the first time.
Agents had already moved in
The goal of agent-created orgs is not to open the door to agents in Axiom — they've been here since we launched our MCP server in 2024. Most queries in Axiom are run by agents now. Our customers have agents create their dashboards, manage their monitoring stack. When there's an incident, agents go to Axiom for the context they need to debug the issue. The only thing agents couldn't do was sign up.
This wasn't an issue 12 months ago, when frontier models still struggled with multi-hour tasks. To be ready for how things stand today, and for how things will look in the very near future, we're making sure Axiom is agent-native at every single point of interaction.
We're not the only ones who see it this way. Cloudflare recently launched temporary accounts for AI agents, which let an agent deploy first and find a human owner afterwards. Expect this to become the norm: when your agent picks a service, the ones that stop and wait for a human won't make the shortlist.
Limits (and lack thereof)
An agent-created org has the same ingest limits as Axiom's existing Personal tier, pro-rated over its 24-hour lifespan. Agents can use it to ingest and query real data, create dashboards, and build monitors. The only new limitation is on notifiers attached to monitors — while agents can still create and configure them so that they're ready on handover, they won't actually fire until someone claims the org. This keeps you safe from spam, while still making it possible for the agent to handle every step of org setup.
The handover
When you go to claim an agent-created org, you can first review what your agent has been up to — how many datasets were created and the total ingest to date. Sign in (no account yet is fine; claiming creates one) and the org is yours. You'll start on Axiom's Personal plan, which comes with an Always Free allowance of 500GB of ingest per month. If that's the right size for you, it's yours to keep. When your usage outgrows it, or you want to invite team members to collaborate, upgrading to Axiom Cloud is just a click away. You only pay for what you use, and usage limits are available to cap your monthly spend.
Get started
If we get this right, you don't have to do anything to 'Get Started' with Axiom. Just be more ambitious with the next task you're passing off to an agent. If you're starting a new project, demand that observability is set up. That dashboards are in place to track key metrics, and that monitors are configured to alert on any performance degradation. Demand a solution that can handle any load, and that rewards scale with even better economics. Your agent will find us, and we'll be ready.
Or if you don't want to wait: