Who's reading all this AI writing?

Hi {{first_name|friend}},

More of what lands in our inboxes at work is written by AI now, and this week one of Microsoft's top executives admitted he had his own AI read it for him. My fix for that is below, and you can try it in 10 minutes.

My Take

Someone sent a Microsoft EVP what was clearly an AI generated doc. He read it, realized it was all AI and had his own AI summarize it. So AI is now being used to push back on AI slop. There's an irony in there somewhere ;)

Ryan Roslansky's own words: he "ended up with nothing that brought new thinking to the idea… This is a doom loop."

If slop grenades are showing up inside the very tech companies selling AI, there's a larger problem impacting everyone's work. When it takes longer to untangle and understand a verbose AI document than it did to generate it, we're in a world of hurt.

And the cost goes past wasted time. Companies are paying for AI tools so people can produce more of this, and some are starting to pull back. Meta is scaling back plans to replace staff with AI agents. The head of the Society for Human Resource Management summed it up: "we're driving the truck and building it at the same time, and that's a problem."

A lot of companies handed everyone an AI tool and skipped the part where someone writes down what good work looks like. A small business can make the same mistake, with fewer people around to catch it.

That missing step is where I think our experience pays off. It's hard to describe what a good proposal or a good client email looks like if you've never written hundreds of them. If you've done the work for 20 or 30 years, you can put it in a few plain sentences, and your AI can check its draft against your standard before anyone else sees it.

As I teach in my classes, human grade output starts with a process:

1. Write down how the task is done and turn it into a Skill. QA it thoroughly before you trust it.

2. Always review the output before sending it to someone else.

3. Fix what's broken, then improve the Skill for the next run.

Here's what step 1 can look like in a small business. Say the task is the follow up email after a client call. Your written standard can be a few lines: under 150 words, confirms what you agreed on, names one next step with a date and sounds like you What’s broken is anything you'd have to rewrite before sending. Put that in a Skill and every draft starts from your standard. When one output still misses, fix the miss in the Skill instead of only fixing the email.

If you only fix the doc, you'll be editing the same thing again next week.

Brett

Try This Today

Write down what good looks like for one task, then have AI check its own draft against it. It takes about 10 minutes.

1. Pick one thing you'll draft with AI this week, like a client email or a proposal.

2. Write 2 or 3 plain sentences on what a good one does and what a miss looks like.

3. Paste your sentences and the AI draft into ChatGPT, Claude or Gemini with: "Here's my standard for this task. Check the draft against it and list every place it falls short. Don't rewrite it yet."

4. Fix what it finds. Save your sentences in a doc so you can paste them in next time.

Watch: why I think we're going numb to how fast AI is moving

@aiover50

I’m thinking that we’re becoming numb to how quickly AI is advancing compared to past tech waves…are you feeling the same? #aitipsandtric... See more

Worth Knowing

Meta's Muse AI is catching on with older adults, worrying privacy experts

Meta's new AI assistant, Muse, is drawing an older crowd. One analytics firm estimates 46% of its US users are 46 or older. An 82 year old in New Jersey used it to compare Medicare drug plans and found one that could save her over $6,000 a year and she checked its work before trusting it. Privacy experts worry the assistant's friendly design makes it easy to share more than you meant to, and security researchers have already found a flaw that could let an attacker reach user data.

The advice from Michael Stokes, a cybersecurity executive quoted in the piece, works for any AI assistant: "Connect only the accounts a task requires, and disconnect them when the job is done." Read more

Banks are finding AI agents inside customer accounts

Major US banks are finding AI agents logged into customer accounts. Usually the customer gave the agent their username and password, which most banks don't allow. Some agents have called bank call centers or taken steps the customer didn't ask for. Morgan Stanley has started blocking them, JPMorgan, Bank of America and Citi are restricting them, and most banks are moving toward read only access, the kind budgeting apps like Quicken already get.

Chase's advice is to never give your username and password to anyone, and that includes an AI tool. If an AI app asks for your bank login, that's your cue to stop. Read more

A McKinsey partner says AI fluency is the next skill every worker needs

Maurice Obeid, a senior partner at McKinsey, argues AI fluency is the next broad skill shift, the way spreadsheets once were. He means using AI in daily work, judging what it produces and knowing what to verify. Job postings that mention AI skills rose 60% in the first half of 2026, and an earlier analysis found jobs asking for AI skills pay about 28% more. His main point: "People do not become fluent by learning about AI. They become fluent by using it, every day, in their real work."

Judging the output and knowing what to check is the part of fluency that leans on experience, and you and I have plenty of that. Read more

Tell me

Have you been on the receiving end of one of these long AI docs yet? One word is fine, I read every reply.