Agent Upstart

Cool things people are building with agents — and practical help getting the most out of them.

Playbooks · Everyday tasks

Hatz AI 2.0: chat that plans, browses and saves automations

Written November 13, 2026

Hatz AI 2.0 turns chat into an agent that plans, browses and saves automations for reuse. What the launch material actually promises.

What changed

Hatz's changelog entry for 2.0 describes a new engine behind its chat: Hatz now plans each task, splits work across helpers acting at the same time, works in the apps and websites the task needs, and saves what works as an automation you can run again. That is the whole pitch in one line — the chat is no longer only answering; it is laying out steps, doing them, and keeping the result as something reusable.

Hatz is a platform sold to small and mid-sized businesses through managed service providers, so the audience for this change is not developers wiring up an agent framework. It is a person at a small company who describes a job in chat — a weekly client report, a ticket triage — and gets a plan, a run, and then a saved automation.

One Auto mode

The older Lite, Performance and Turbo modes collapse into one Auto mode. Auto picks the model for each step of a task and runs parts of the work in parallel with helpers. Before it starts, it lays out the steps and shows each one as it runs; you can edit the steps to add, reorder or remove them. When a task needs a decision, Auto asks a few questions up front and marks a recommended answer for each. When a task needs an app that is not connected yet, Hatz offers to connect it in the chat and then carries on.

Hatz publishes its own benchmark figures for Auto, from internal testing on what it calls real MSP and SMB use cases: 24% better outputs than Turbo, 15% more cost effective than Performance, and 12% faster than Lite. In head-to-head testing, Hatz says Auto used 68% fewer credits than Sonnet 5.5 for the same tasks and returned results 76% faster than Terra 6.1. Those are Hatz's numbers, from its own benchmark; the changelog does not publish the tasks or the method behind them.

Automations and browser work

Automations are repeatable tasks you build by describing them in chat. When a chat does something you will need again, you save it as an automation, then run it on a schedule or run it on demand with new inputs. Unlike Hatz's workflows, automations are not built step by step in an editor: Hatz builds the automation from the work it just did in chat, you edit it by chatting, and opening one shows its steps and what each step does. Existing workflows can be converted: Auto shows a preview of each one, and only the ones you select are converted.

A new Browser Control lets Hatz open websites in a secure browser and work in them, including vendor portals, billing sites and tools without an integration. You can watch the browser live next to the chat. If a site needs a person — a sign-in step, for example — Hatz pauses, hands control to you, and picks up where you left off when you hand it back.

Running several jobs at once, and keeping keys out of chat

Parallel Chats let you run as many chats as you want at once. Each chat can keep working for up to three hours, even after you leave it, and the sidebar marks a chat that needs you. You can steer a running chat with added instructions, or ask a side question without changing the task. Approval settings decide how much runs unattended: Smart runs low-risk actions on its own and asks before anything risky, Manual asks before more actions, and Stop ends a run at any point and keeps what is finished.

A Vault gives API keys a home outside the chat. You add a key once, with a name and the sites it is for; when a task needs it, Hatz attaches the key itself, so the model never sees it. Only you can use your keys, and you can edit or stop using any of them. For administrators, default model settings now offer Auto plus specific models, the older Auto tiers are retired, existing role limits still apply, and rate limiting defaults to Auto.

Limits

Related playbooks

Sources

Back to: Everyday tasks playbooks hub · Also: Claude Duet: Anthropic's proactive agent app