Lab — July 14, 2026
In Today's Newsletter
We Tested Every AI Notetaker Gadget. The $129 Price Is a Trap.
Claude Tag Joined Your Slack. Nobody Invited It.
This $17 Email AI Claims to Clone Your Writing Voice
What else happened today?What AI tools should I be using?

Good Morning Thorium Valley, welcome back to The Lab.

Today we're testing those pocket-sized AI recorders that promise to kill the awkward meeting bot, and finding out if the $129 price is actually what you end up paying.

We're also looking at an AI agent that just shows up in your Slack and starts answering questions nobody asked it.

And we found a $17 email tool that says it can learn how you write and start drafting replies for you automatically.

Quickly before we dive in — When an AI tool gives you a solid result, what's your first instinct?

We Tested Every AI Notetaker Gadget. The $129 Price Is a Trap.

PRODUCTIVITY

We Tested Every AI Notetaker Gadget. The $129 Price Is a Trap.
Share X in

A new wave of pocket-sized AI recorders promises to kill the most annoying part of meeting AI: the bot that joins your call and makes everyone uncomfortable. Drop a device on the table, press a button, and get a transcript and summary — no bot, no lobby, no awkward "is that thing recording?" moment. It's a real problem being solved: Microsoft is quarantining external AI bots in meeting lobbies starting in 2026, and plenty of professionals refuse to speak freely until the notetaker is shut off.

WIRED tested four of these devices, ranging from $129 to $189, and named the Comulytic Note Pro the best of the bunch. It's tiny, records for 45 hours on a charge, supports 113 languages, and produced the most insightful summaries of the group — pulling relevant quotes and catching details the others missed. The runners-up each had deal-breakers: the Open Vision Pocket's single button misfires constantly, the InnAIO T10 is really a translator moonlighting as a notetaker, and the OSO AI Earbuds only last four hours before the app crashes if you accidentally close it.

Now the catch. All four devices offer some free functionality, but the advanced AI features require subscriptions that quietly change the math:

+ The Comulytic Premium plan runs $14.99/month after a three-month trial. Over a year, that $129 purchase becomes roughly $264. Over three years, you're past $600.

+ Competitors are worse: the Pocket is $20/month, the InnAIO is $25/month.

+ Comulytic's free Starter plan includes unlimited transcription and basic summaries, but limits you to ten instant summaries and three detailed analyses a month.

And privacy isn't fully solved either. These recorders store audio locally, but when you use the AI features, your conversations are processed in the cloud by providers like OpenAI or Google. You're removing the bot from the meeting room, but your words still end up on someone else's server. Some HR leaders argue the organizational risks mean companies shouldn't use them at all.

The Verdict

If you do a lot of in-person meetings and the meeting-bot friction is killing you, the Comulytic Note Pro is the best option at the lowest total cost. Just know what you're buying. It's not just a $129 gadget. Advanced AI features require an ongoing subscription, although basic transcription and summaries are free for life. And if most of your meetings are on Zoom or Teams, a software tool like Granola or Fathom can handle the job and often offers free tiers that may cover your needs without adding another device to your stack.

Claude Tag Joined Your Slack. Nobody Invited It.

PRODUCTIVITY

Claude Tag Joined Your Slack. Nobody Invited It.
Share X in

Claude Tag is Anthropic's new AI agent for Slack, and it doesn't wait for you to ask it anything. Add it to a channel and it starts watching conversations, jumping in when it thinks it can help, and remembering instructions across sessions. It runs on Opus 4.8, Anthropic's most capable model — but unlike opening Claude in a browser, Tag is designed to feel like a persistent team member who happens to be online 24 hours a day.

It's in beta for Claude Enterprise and Team customers. Slack only for now, with Teams and email planned next. Here's what it actually does:

+ If someone posts a data question, Tag takes a first pass at answering it

+ If a bug report lands, it can open a pull request

+ Tell it to monitor for a specific type of issue and it'll remember that instruction over time

+ You can adjust its behavior or routines at any time, and it retains those preferences

The pitch makes the most sense for small teams that don't have a dedicated data analyst or someone tracking every channel for stalled requests — capabilities they could never afford to hire for. But the big unknown is whether a proactive AI in your work channels actually helps or just adds noise. An agent that surfaces the right thing at the right moment is a gift; one that interjects where it wasn't needed is a nuisance. Early signals suggest Anthropic tuned Tag to stay quiet more than it speaks.

The privacy angle matters. Tag reads your Slack messages on Anthropic's servers, and because it uses token-based billing, any proactive processing it does in your channels counts toward your usage. If sensitive information flows through those channels, think that through before adding it. It's not the only workspace agent in the works — OpenAI is building its own and Microsoft Copilot already lives inside Teams — but Tag is one of the first explicitly designed to act without being asked.

The Verdict

If your team lives in Slack and you're already on a Claude Enterprise or Team plan, Tag is worth trying, especially for channels with repetitive workflows or constant data questions. If you're on a smaller plan or your team uses Teams, there's nothing here for you yet. And if the idea of an AI reading your channels and deciding when to speak makes you uneasy, that's a perfectly fair reason to wait.

This $17 Email AI Claims to Clone Your Writing Voice

PRODUCTIVITY

This $17 Email AI Claims to Clone Your Writing Voice
Share X in

Agentys is an email AI assistant that reads your sent mail, builds a model of how you write, and starts auto-drafting replies without you prompting it. It costs $16.99 a month, works on top of Gmail and Outlook, and doesn't require switching email clients.

The pitch is simple: instead of opening an email and telling an AI what to say, Agentys watches your inbox and has a reply ready before you get there. It picks up on how you sign off, whether you write shorter sentences to longtime clients versus new contacts, and vocabulary specific to your field. The company says accuracy improves over the first two to three weeks, and early adopter data claims users save about an hour and 47 minutes per day.

That number is worth flagging — it comes from Agentys itself, not an independent study, and the tool is new enough that there's almost no outside coverage verifying whether the voice matching actually holds up.

What's clear is where Agentys fits. Voice-matched email drafting is now a real category, and most options cost more or require you to prompt every draft manually:

+ Superhuman added per-contact voice matching after Grammarly acquired it for roughly $825 million, but you need the $30–40/month Business plan to get it

+ Shortwave's Ghostwriter learns from your sent folder starting at $24/month, though it only works with Gmail

+ Gemini in Google Workspace can draft replies for free but doesn't learn your voice and requires a prompt every time

+ SaneBox handles inbox filtering for $7/month if drafting isn't what you need

Agentys is the cheapest dedicated option and the only one combining voice learning with fully automatic drafting at that price. No prompting, no client switch. For people who answer dozens of emails a day, that per-email prompting cost with tools like Copilot or Gemini adds up fast.

The Verdict

Agentys makes the most interesting promise in AI email right now: learn your voice, draft replies automatically, $17 a month. If that works as described, it's a genuine time saver. The catch is that almost nobody outside the company has verified those claims yet. Use the 7-day trial on your real inbox and judge the drafts yourself. If you want something proven, Superhuman and Shortwave both offer voice matching with much larger user bases and more public track records. They just cost more. And if you're fine prompting each draft manually, Gemini is already free in your Gmail.

In Other News

EVERYTHING ELSE IN AI

What else happened today?

+ Apple sues OpenAI for stealing hardware secrets "at every level," naming OpenAI's chief hardware officer — a former Apple VP — in a 41-page complaint

+ Oracle cut 21,000 jobs and blamed AI in a regulatory filing, becoming the first major company to formally name AI as the cause in an annual report

+ Delaware proposes a legal entity for AI agents so autonomous software can sign contracts, get sued, and be held accountable

+ OpenAI is in early talks to give the U.S. government a 5% stake worth roughly $42.6 billion

+ Microsoft spins out a $2.5B AI consulting firm that will help enterprises adopt AI from any vendor — even Microsoft's competitors

+ Saudi Aramco leads an $800M bet on open-source AI, backing Together AI at an $8.3 billion valuation

+ Video-generation startup PixVerse crosses $2B valuation after raising $439M with Alibaba joining the round

+ Meta boosted its Louisiana AI data center budget to $50B+, up from $10B when announced two years ago

AI Tools

OTHER TOOLS

What our editors are paying attention to today

+ Wispr Flow: *(sponsored)*: The new viral voice-to-text AI for iPhone and Mac that actually understands what you're saying every time

+ Superpower: *(sponsored)*: The health app that tests your blood and uses AI to tell you what's actually going on with your body before your doctor does

+ ChatGPT Work: OpenAI's new agent takes a goal — like "build me a competitive analysis deck" — and works across your apps for hours on its own, returning finished documents, spreadsheets, and presentations instead of chat replies

+ Claude Reflect: Anthropic added a Spotify Wrapped-style dashboard that shows your Claude habits — peak hours, most-used tasks, repeat requests — so you can see exactly how you've been using AI over the past year

+ LALAL.AI Lynx: A new neural network that cleanly separates vocals from any audio track — useful for podcasters cleaning up interviews, musicians isolating samples, or anyone trying to strip background noise from a recording

+ ChatCut: A Codex plugin that turns OpenAI's coding agent into a video editor — describe the cut you want in plain English and it removes silences, adds captions, and reframes shots, then hands you an editable timeline instead of a locked file

+ macOS Golden Gate: Apple's public beta is live with the fully revamped Siri AI that can read your screen, answer questions about what you're looking at, and take actions across apps from a single voice command

That's the Lab for this week. If a tool in here saved you time or wasted it, tell us — reply directly.

Written by Jason Chen, Advait Prakash, Andrew Hales, and the Thorium Valley crew.

That's all for today's Lab. See you next time.