The AI note-taking race just flipped

PRODUCTIVITY

Five weeks ago we recommended Granola as the best meeting notetaker for people who don't want a bot joining their calls. Then the company raised $125 million, capped free notes at 30 days, and quietly made shared notes viewable to anyone with a link by default. We owe you an update.

Granola's pitch was always simple: it listens through your system audio, no bot pops up in the attendee list, and it blends your typed notes with what was actually said. The free tier was generous enough that most people never needed to pay. That tier still exists on paper, but anything older than 30 days now gets locked behind the Business plan ($14/user/month) or Individual plan ($18/month). Need to reference what a client agreed to two months ago? That note is behind a wall. Worse, shared note links are viewable to anyone who has them, and your data gets used for internal AI training unless you manually opt out. For a tool that sold itself on privacy, those defaults undercut the whole premise.

Meanwhile, the competition caught up:

+ Fathom shipped bot-free capture with full speaker attribution on its free tier — unlimited recordings, no 30-day cutoff. It also took a direct shot at Granola for not identifying who said what.

+ Otter remains strong for teams wanting full automation at $16.99/month.

+ tl;dv offers unlimited free recordings if you just need clips to share.

Granola's CEO has said he never would have entered this space just to generate meeting notes — which tells you where the product is headed. Granola earned SOC 2 certification and is clearly building for enterprise deals, not the solo user who loved the free plan. With $125 million in VC funding, the roadmap now answers to growth targets.

If your company needs SOC 2 compliance and you're buying seats in bulk, Granola at $14 per user per month is still a solid product with a genuinely clever approach to note-taking. Everyone else should switch to Fathom. It now matches Granola's core advantage of bot-free recording, it identifies speakers properly, and its free tier doesn't hold your meeting history hostage. The tool we recommended five weeks ago still works well. It just stopped being the best deal, and the privacy defaults made it harder to trust.

SEARCH

Analysts estimate Google earns roughly $300 a year in ad revenue per active user. Kagi charges $10 a month and shows you nothing but results.

No ads, no tracking, no AI summaries unless you ask for one. You pay, you search, you're the customer instead of the product. As founder Vladimir Prelovac put it: "For Kagi, user data is a liability. We don't want it. We have nothing useful to do with it."

The pitch sounds absurd until you actually use it. Results are clean — no sponsored listings, no AI Overview you didn't ask for, no SEO-gamed listicles clogging up page one. Kagi also gives you controls Google never would:

+ Domain ranking: Block or boost specific sites, so if you never want to see Pinterest again, it's gone permanently

+ Lenses: Filter results by category — academic papers, forums, news — so you skip the noise before it starts

+ Quick Answer: An opt-in AI summary you explicitly request, otherwise it stays out of the way

Prelovac says Kagi users perform about half as many searches as they did on Google because results are more precise. The retention numbers back that up: of the 210 people who signed up on launch day in June 2022, 121 are still paying three years later — 57% retention for a product most people assume should be free.

The counterarguments are real. A subscription search engine called Neeva tried a similar model with former senior Google executives and more funding, and still shut down after concluding paid consumer search wasn't viable. Kagi is also reportedly considering pulling its AI assistant from the $10 plan to manage inference costs, which some users see as a bait-and-switch.

So why hasn't Kagi followed Neeva? No major VC pressure, just $2.5 million in outside investment plus $3 million from Prelovac himself, and the company hit break-even profitability in mid-2024 with 50,000 paying members and 45 employees. It's a Public Benefit Corporation with a stated goal of reaching 5 million subscribers with fewer than 75 people — a fundamentally different bet than trying to outscale Google.

If you search constantly for work and Google's ad clutter genuinely costs you time, Kagi at $10 a month pays for itself in focus alone. If you mostly need AI-powered research, Claude or Perplexity will serve you better. And if you just want privacy without paying, Brave Search is free with its own independent index. Kagi is for the person who treats search like a tool, not a habit, and wants it to work on the first click.

PRODUCTIVITY

Superhuman costs $360/year. Spark costs $80/year. One just absorbed an $825M company.

Grammarly didn't just acquire Superhuman last July. It renamed itself Superhuman. The acquirer took the acquired company's name. That almost never happens, and it tells you everything about where the email app market is headed.

Superhuman Mail at $30/month was already a tough sell for most people. It was the fastest email client on the market, built around keyboard shortcuts and speed, and the people who loved it were borderline religious about it. The killer feature is Auto Drafts — you open an email and a usable reply is already sitting there. That's why 94% of weekly active users actually use the AI features, an absurdly high adoption rate for any software tool. But Auto Drafts now requires the $40/month Business plan, and there's still no free tier.

Spark is the opposite bet. The free tier is genuinely good — it sorts your inbox, works on every platform including Windows, and supports Gmail, Outlook, iCloud, and anything with IMAP. The Plus plan at $10/month includes AI compose and rewrite, and teams get shared inboxes and collaborative drafting at a fraction of what enterprise tools charge. The catch: Spark's AI is thin. It doesn't learn your voice, doesn't triage your inbox, and doesn't draft replies proactively. Users consistently describe it as comparable to pasting an email into ChatGPT.

What's changed is that Superhuman isn't really competing with Spark anymore. Grammarly brought a much bigger arsenal to the table:

+ 30 million daily active users and a mature writing engine

+ An AI detector after acquiring GPTZero

+ A cross-app agent called Go that works across your browser

That's a platform pitch, not an email pitch. Meanwhile, Notion Mail just shut down entirely. The standalone premium email category is shrinking fast.

If you process 100 plus emails a day and speed is worth $40 a month to you, Superhuman on the Business plan is still the best email experience available, and the Grammarly platform building around it only makes it stickier over time. If you need team email on a budget, Spark's Pro plan does the job at a quarter of the price. Just don't buy Spark for the AI. Buy it for the shared inboxes and the fact that it works everywhere. And if you're on Gmail and want better AI writing for less, Shortwave at $7 a month is worth a look before you commit to either.

EVERYTHING ELSE IN AI

+ Meta plans to sell its spare AI computing power to rival AWS, Azure, and Google Cloud — stock jumped 10%

+ US lifts export ban on Anthropic's most powerful AI models just three weeks after shutting them down over national security fears

+ Netflix will use AI to recreate Gene Wilder's voice for a new Wonka reality show — fans are not having it

+ Nvidia's most ambitious AI chip has been quietly cut in half after TSMC hit a packaging ceiling it can't solve until 2028

+ Ford, IBM, and others are rehiring human workers after discovering AI couldn't actually replace them

+ Together AI raises $800M as open-source model infrastructure hits $1.15B in annual bookings

+ Singapore seizes a $42 million luxury home tied to Nvidia AI chip smuggling ring

+ Palantir's CEO says token-based AI pricing is broken calls out OpenAI and Anthropic by name

OTHER TOOLS

+ 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

+ Grok Voice Agent Builder: xAI launched a no-code platform that lets anyone build human-like voice agents for $0.05 per minute — no need to stitch together multiple services or write a single line of code

+ GitHub Copilot: GitHub added Kimi K2.7 Code to Copilot's model picker, marking the first time an open-weight model is available as a selectable option — giving developers a lower-cost alternative for coding workflows

+ Hostinger Ecommerce: A new platform that lets small sellers snap a product photo and get a working checkout link in minutes, with one dashboard connecting to websites, TikTok shops, and messaging apps without rebuilding anything

+ Google BigQuery: Conversational Analytics is now generally available, letting anyone on a team ask business questions in plain English and get multi-step analyses and visual reports back — no SQL required

+ Engram: A new startup backed by $98M and angels like Andrej Karpathy just launched a memory layer that trains AI to actually learn your company's knowledge in advance, so it gives expert-level answers using up to 100x fewer tokens