Your AI is only as good as what you feed it. Feed it a transcript with your product names, acronyms, and numbers spelled wrong, and you spend the afternoon hand-fixing the output.

Wispr Flow Notetaker uses your dictionary and calendar, so product names, acronyms, numbers, and uncommon names come out spelled right. It pops up when your meeting starts and captures Zoom, Google Meet, Teams, and Slack huddles with one click, no bot joining the call.

Then your meetings show up inside Claude or ChatGPT through the built-in connector. No copy paste. Try Notetaker free on Mac and Windows.

Try Notetaker Free

Claude Duet and the morning update.

Yesterday, Anthropic launched Claude Duet, a new AI agent built for professionals. Its centerpiece is a morning brief.  The Verge frames Claude Duet as Anthropic taking on Muse, (a name we are naturally fond of around here!). The facts so far are simple: an agent, a professional audience, and a daily digest that arrives before you have chosen what to care about that day.

Every brief is an act of editorial judgment. Someone, or something, decides which of the 200 overnight messages you receive is worthy of a single line, and which ones vanish. Media scholars have understood this for decades. In 1963, Bernard Cohen observed that the press may not be very good at telling people what to think, but it is stunningly good at telling them what to think about.  A newspaper editor chooses the front page; now an agent chooses the top of your day.

The key takeaway: compression is judgment. A model deciding “what matters” draws upon previously established patterns of what usually matters: the deadline, the executive's name, the urgent subject line.  As we know, AI hones toward the mean, the average…the mediocre.  That works well for the typical exec.  The best professionals, though, are rarely typical.  The email that changes your quarter might be a vague note from a colleague you rarely hear from, or a question phrased so politely that only a skilled human reader detects the underlying sentiment.  Our human minds are pattern detectors, tuned to our own human experience, fears, and hunches. A brief tuned to “the average” is still not tuned to any specific job role or industry.

The digital agent (whether that’s GrokBot, Duet, dots, or Muse) takes aim at the role of the “chief of staff,” who does much the same thing. Executives have relied on them for a century.  However, a human chief of staff learns your priorities through years of friction, and you can debate with him or her regarding why they chose to exclude something.  With an agent, the onus is on the human to sketch that understanding out more deliberately.

Three habits are worth adopting from day one:

First, write your own top three priorities before you read the brief, so the agent's framing meets a plan that already suits your own goals. Second, skim the raw inbox once a week to see what the agent’s summary consistently excludes. Third, ask the agent to list what it excluded and why (my personal favorite). Details of how Duet chooses what to surface are what we’ll be watching as more professionals use it. The tool that lets you audit its choices can earn more trust than the one that simply sounds confident.

also on the radar.

an agent that removed its own monitor.

OpenAI says an AI agent disabled its safety monitoring system and then misreported what it had done, The Verge reports. The agent was pursuing a task, and the monitor was an obstacle to that task, which is a reminder that these systems optimize for the goal in front of them and have no instinct for the contextual rules. Human review continues to be critical here, and it works only if someone is reviewing the logs, rather than the agent's own account of them.

Walmart lets agents shop, within limits.

Walmart is inviting AI agents to shop on customers' behalf, and the design includes spending limits and approval gates. The human shopper still authors the intent and sets the budget, while the agent handles the legwork, a division of labor that looks sensible and teachable.  As I’ve previously reported, Amazon has taken the opposite stance (shutting out agents like Meta’s Muse), so the two largest retailers are now running a live experiment in whether shoppers want agents at the door or are kept outside it.  FYI, Shopify, like Walmart, has taken a more permissive stance toward personal AI agents, like Muse, through its “agentic checkout.”

the ad-funded AI your doctor may be using.

Ars Technica profiles an AI search startup, called OpenEvidence, valued at $15 billion (and funded by advertising).  They report that nearly 66% of US doctors consult an AI search engine when making patient decisions. Clinicians have always relied on references, but a reference supported by advertisers raises a fair question about whose answer influences your doctor’s recommendations. Patients may soon want to ask not only what their doctor consulted, but who paid for that tool to exist.  And, I might add: if the doctor is using the AI tool, why shouldn’t the patient use the AI tool directly and cut out the middleman?

worth your time.

Herbert Simon's 1971 essay "Designing Organizations for an Information-Rich World" contains a line that reads like a forecast for this week's lead story: "a wealth of information creates a poverty of attention." Simon wrote it decades before language models, and his argument is that the scarce resource in any system is not data, but the human attention to process it, so the real design question is what to filter out and who decides.   This is increasingly an issue throughout our AI-powered society, where AI generated research, AI-generated small claims court filings, AI personal agents, are flooding systems meant for human processing and interaction.  Check it out with your morning brief open in another tab. It will sharpen how you think about every agent that promises to save you time by choosing for you.

The human mind is the original generative machine.