Tool Radar: A New Flagship Model, and Agents That Act

Welcome back to Tool Radar — our running note on the adjacent tools we are watching for freelancers. By "adjacent" we mean the tools that sit next to the work itself: the AI assistant you draft with, the design app you mock up in, the editor you ship from, the notes app that holds your second brain. We do not cover other freelance-management apps here — this is about the rest of your stack.

Two filters run through everything below. The first is the question this series keeps returning to: does it remove a decision, or does it just add a tab? The second is the one solo freelancers forget until it matters: what happens to your work if this tool disappears — can you get your stuff out? This was a busy week, and both filters earned their keep.

AI: a flagship went public — and started doing whole jobs

The headline is that OpenAI released GPT-5.6 to the public, in a three-tier lineup called Sol, Terra, and Luna — a flagship, an everyday-work option, and a faster, lighter one — after a stretch of limited, review-gated access. Alongside it came ChatGPT Work, an agent pitched at carrying out longer, multi-step tasks rather than just answering a question.

We will describe this soberly, because that is more useful than the launch-day noise. A new flagship model means the first draft you can pull out of an assistant is more convincing than last month's. That is genuinely helpful — and it is exactly where the risk hides.

Where we would stay steady:

• A stronger draft is still a draft, never a send. For anything a client sees — a proposal, a scope description, a payment-reminder email — read every line yourself. A better model writes a more persuasive wrong answer just as fluently as a right one. The bar for editing goes up, not down.
• "Does a whole job" is where the first filter bites hardest. An agent that runs several steps on its own can genuinely remove a decision — or it can add a tab full of work you now have to inspect, correct, and often redo, because the client is paying for your judgment, not the tool's first pass.
• Keep the output, not the account. Whatever you draft in, paste the result into a document you own. Model names and tiers change constantly; your proposal and your scope language should live somewhere that does not depend on which one is current this month.

This is also how we think about AI inside Drift Catch. Our two AI features — AI-Assisted Drafting on templates, and support-email triage — both produce a draft that a human reviews before anything is sent. We let the model get a first version onto the page; we never let it have the last word.

Design: when a layer becomes live code

For visual freelancers, the item to watch is Figma's Code Layers, announced at Config 2026 and rolling out to a waitlist beta starting this month. The pitch: turn any design layer into an interactive, code-backed layer with a click or a prompt, riff on it beside your other frames, and — for the more technical — clone a repository onto the canvas and sync changes back.

Run it through the first filter honestly. Sometimes collapsing "design it, then hand it to code" into one surface removes a decision — the tedious translation step you would have babysat anyway. But a live code layer can also add a tab: another moving part to maintain, in a file that now behaves less like a picture and more like software.

A tool that blends two crafts saves you a handoff only if you were going to do both crafts. If you were not, it is one more thing to keep working.

The second filter matters here too. The reassuring part is that the code side can live in a repository you own, which is about as portable as it gets. The part to watch is the design side: the more your finished work lives as an app-specific project only that tool fully understands, the more it matters that you can still export a flat, openable file — in case the tool, the tier, or the beta changes under you.

Productivity: agents you assign like a teammate

In the notes-and-projects lane, Notion shipped a release that lets external agents — the ones living in your editor, CLI, or a separate app — be assigned tasks from a shared board and @-mentioned like a teammate, plus interactive blocks its own agent can build inside a page. Assistants such as Claude and Cursor were named among the first that plug in this way.

This is the first filter at full volume. An agent you can hand a task and watch run is a real step past "another dashboard to check" — if the task is one where a wrong result is cheap to catch. On a client deliverable, where a subtle miss ships quietly, we would keep a human between the agent and the send, the same way we do with our own drafting.

And the second filter never stops applying to the app that holds your second brain. Before you let more of your working life pool inside any single workspace, confirm you can export it into a plain, readable format you own. The more capable the workspace, the easier it is to forget you are renting the exit.

General information, not legal or financial advice — your own situation may call for something different.

What this has to do with us

Most of these tools help you do the work. The thing almost none of them touch is the agreement around the work — what was promised, what counts as extra, and when you get paid. That gap is the whole reason Drift Catch exists, and it is the one part of your stack we would argue you most want to own outright rather than rent loosely.

If you are weighing any of this week's launches, our two filters are a decent first checklist, and our note on how to choose freelance management software goes deeper on the same instinct. If "can I get my data out" is the thread that pulled you through this edition, privacy-first freelance software is the matching read. And if you would rather the whole signed-to-paid flow lived in one place, that is what our all-in-one software for freelancers is for.

We run Tool Radar regularly. If there is an adjacent tool you think every freelancer should know about, tell us at ideas@driftcatch.app — we read every message, and good suggestions tend to show up in the next edition.