TechCrunch Arrives, And So Does The Permissions Conversation

What people were saying

Today was the day the discourse got a press peg. @TechCrunch framed it as early testers raving while others squint at the sweeping access, broad terms, and the whole acting-on-your-behalf premise. It is a fair framing, and it produced the exact reply thread you would expect: some people alarmed, some people pointing out the product is still in closed beta with a privacy policy that is, at the moment, a bare HTML page, and at least one person delighted to have been quoted in TechCrunch at all.

The most substantive pushback came from @startupandrew, who runs a competing agent product and made the case that an agent that YOLOs it can produce great demos, while an agent you would trust with real data and real consequences is a much harder engineering problem. He is not a neutral party here, and he did not pretend to be. But the argument landed: one Instinct user in the replies said an earlier version of the same comment had already gotten them to sign up for the alternative.

The praise

On the other side of the ledger, @PositiveGamma said Instinct is now pinned to the top of WhatsApp, and that queries which would have gone to Google a year ago and Claude or ChatGPT last month are now going here instead. The specific thing he liked was less the flashy task execution and more the boring part: staying on top of scattered WhatsApp threads and plans.

@ay_ushr made the payments case, arguing he is fine handing an agent his card as long as the confirmation mechanisms are good, and that he cannot name a use case that existing card rails do not already handle. This was in response to a broader point that agentic payments are already happening quietly through computer use and logged-in CLIs, which is, when you think about it, exactly what Instinct is.

The pushback

The objections clustered around permission design rather than capability. Nobody today argued the thing does not work. They argued that broad access plus autonomy plus a thin policy is a bad combination, that logs and kill switches and reversible actions should be visible product surfaces, and that regulators may want a look. One reply put the fear economically: give it sweeping access, get a great demo, find out the rest in three months.

Off to the side, @toptickcrypto suggested the real opening is a verifiably private version built on TEEs and open-weight models, which is either the correct long-term answer or a very elegant way to describe a product nobody has shipped.