Review: an email assistant inside Gmail, AI that edits whole designs, and why your analytics now need to know who is human

This week's launches share one idea: AI is moving into the tools your team already works in, instead of asking people to come to a separate chat window. As always, these are first assessments from what has been published, not test results.

The reviews

Superhuman's Email Assistant in Gmail and Outlook. Verdict: worth a trial for anyone whose inbox runs their day. It separates what needs a reply from what is just for information, brings important conversations back to the top, archives low-priority mail, and drafts replies in your voice using context from other emails and online research. The part I like is that users stay in control of how it writes, schedules and triages. For a marketing lead juggling agencies, vendors and internal requests, sorting alone could save real time. Keep drafting on a short leash at first: a reply in your voice that gets a fact wrong is worse than a slow reply.

Managed connectors for Claude Enterprise. Verdict: the setup most teams should have had from the start. Administrators can now switch on connectors to the company's apps centrally, decide which groups or roles get them and at what access level, and people get them automatically the first time they log in. When every person connects tools on their own, nobody knows who can reach what. If your team uses AI with company data, ask for this kind of central control, whatever tool you use.

Figma's Google Drive connector. Verdict: useful for teams whose briefs live in Drive. Figma's agent and Make can now pull research, data and images from Google Drive into a design conversation and create new Drive files. It reaches every plan, including Starter. The risk is the usual one: an agent will use whatever it finds, so keep the brief folder clean and current.

Canva's Brush model. Verdict: promising, wait for Brand Kit support. Canva says Brush edits designs up to 69% faster than its previous design agent, at 46% lower cost per task, and reads both content and layout across full designs like multi-page decks. It is in beta for Canva Pro subscribers. Brand Kit support is planned but not here yet, and for most marketing teams that is the feature that decides whether edits stay on brand.

The finding that matters more

The most useful piece this week was about analytics. PostHog says agents now create about half of its dashboards and feature flags, so the team tags every event with whether a human or an agent did it. That tagging showed that a third of organizations query the product through agents, and 1 in 5 new organizations signs up through one.

If agents read your pages, compare your pricing and even start sign-ups, then a funnel built for people will mix two very different audiences. Conversion rates, time on page and drop-off points all change meaning.

A related planning piece made the budget version of the same point: put AI spend in your plan next to the people plan, and show when the extra output will reach the numbers. AI costs that sit in a separate line are easy to question and hard to defend.

What I would do

• Ask your analytics owner whether you can tell human visits from agent visits. If not, start with the traffic you already label as AI-referred.
• Put one AI tool from this list on a two-week trial with a real task, and write down what it saved.
• Add AI spend to next year's plan as its own line, next to the work it supports.

Worth trying

Use precise numbers where lower is better. One study found that precise figures make a feature count for more in buying decisions than rounded ones, but only for features where lower is better, like weight or price. "Loads in 1.8 seconds" may beat "loads in under 2 seconds." Check your landing pages for rounded numbers that could be exact.

AI is showing up inside the tools you already use. The teams that benefit will be the ones whose reporting can still tell who is on the other side.