Personal AI agents want your inbox and your card
You enter your email address on Substack. Unsubscribe at any time.
Personal AI agents are here, and they want your inbox and your card
In the past two months, we've seen a flurry of announcements for personal digital executive assistants. Meta's Muse, OpenAI's Dots in ChatGPT, Anthropic's Claude Duet and newcomer Instinct have all launched, and we reach them where we already are, by text message, on WhatsApp or in an app on our phones. They work across our inbox and apps to book the trip or do the shopping, and since late October Instinct lets eligible US users skip approval on purchases under a limit they set for each one.
They promise to save us an immense amount of time, and in my initial testing, it's true. They are easy to use, and when I interact with Instinct, I do so within iMessage instead of having to install another app. But not everyone is excited about this. Mehdi Jamei of Veris AI told Moneywise that Instinct used a login code from his Gmail to get into an account without asking, then hallucinated an explanation of how it got in. And as Terry Angelos writes for a16z, most wallets treat a connected agent's purchase as authorised even when it misreads you, so it's unclear who makes it right.
As we begin using these personal assistants, let's be mindful that our companies' details could be exposed through our interactions with these agents. Although we know that we'll increase our ability to get things done, both at home and at work, we need to ensure we're not exposing company data or account details while we're still battle-testing these new life-improving bots. For our CIOs, this is the opportunity to get ahead of our team members by clarifying our policies on AI personal assistants before things potentially get out of hand.
Also worth your time
The judgment bottleneck: Balancing AI and human speeds in strategy, McKinsey Quarterly. One executive's strategy team now does five weeks of financial analysis in under two days, but our judgment hasn't sped up, so the review pile lands on our desks and the junior work that trained future leaders disappears. The fix is to sort decisions into three tiers, name one accountable owner for each big call, and teach teams to explain the assumptions behind every AI output.
Transformer, Deployer or Experimenter?, Considered Machines. Paul Willmott reads six 2026 consultancy surveys and finds only 6% to 13% of firms getting significant value from AI. In BCG's survey, 95% of that top group track it through clear KPIs or the P&L, against 53% of firms still experimenting. A great one for the board: agree honestly which group we're in, then fund a multi-year programme with clear stop rules instead of another string of pilots.
Our organizations are being DDoSed from the inside, Karri Saarinen. Linear's CEO argues that teams working with agents now ship changes faster than leaders can review them. Companies are slowly losing their ability to steer the ship. It's a different form of AI slop, where just because we can do something quickly, we do. We forget to ask ourselves, "Is this the right thing to build?" A great read for anyone struggling to keep their products tight and their teams focused on the right things.
LSEG CTO on AI unit economics, innovation, and P&L for tokenomics, The Stack. Ken Beinert, the London Stock Exchange Group's chief technology and operations officer, makes each product's P&L owner responsible for its AI spend, backed by evals and cost controls that enterprise AI and finance run together. And he budgets well beyond tokens, for data preparation, guardrails, recurring evals and monitoring. It's worth asking our CFOs who owns the AI bill, and whether we're budgeting for all of it.
Is your leadership team ready to light the fire?
Upskill with AI Firestarter bootcamps and workshops, built for leaders who are too busy leading to be experimenting with AI.
As a leader, you set the tone for how AI is used in your organisation.
Until next time,
Ryan Snyder
ryan@aifirestarter.com