Twitxy Saved

Chet Kapoor

Verified account This account is verified. Learn more

@ChetKapoor

VP, AWS
Believe, Inspire, Execute
aws.com Joined March 2008
544 Following 6.2K Followers 2.5K Posts

Latest posts 5

Chet Kapoor Verified account @ChetKapoor
Before giving an AI agent more authority, test how it fails.

Have it attempt one unauthorized change among legitimate changes by humans and agents.

Can you stop it? Trace it? Undo only the damage?
The limit of agent autonomy is not intelligence. It is control.
15 replies 61 reposts 707 likes 73K views View on X
Chet Kapoor Verified account @ChetKapoor
The cheaper AI model is often the more expensive one.

Model A: $2/M tokens. 6/10 usable per run. 3 runs, 15 min.

Model B: $4/M tokens. 9/10 usable. 1 run, 3 min.

B costs 2x per token but ~1/7 as much per usable result, counting human time.

Track cost per usable result.
0 replies 1 reposts 11 likes 1.7K views View on X
Chet Kapoor Verified account @ChetKapoor
Every hotel key card stops working at checkout.

Software keys don't. API tokens, agent credentials and other secrets are keys given to machines, and those machines never check out. What if your agent checks itself in at 3AM unauthorized?

As agents multiply, so do these keys. And the newest threats are not from outside in. They just log in with credentials that still work.

The next big push in identity security is about guests who never leave. Give every key limited access and a checkout time.