(no title)
davedx | 5 days ago
Assumptions that I think warrant closer inspection:
- agents will always win vs antibot firewalls: highly doubtful given my experience with openclaw. Antibot measures are everywhere, they're advanced, and the more agents threaten legacy business models the harder they'll fight to protect them. Think e.g. Uber investing more in anti-bot tech to stop agents turning them into a whitelabel API. Think CloudFlare's recent moves in this area. Think Salesforce reducing access to Slack API. Data moats will be guarded more strongly.
- total cost of inference will be cheaper than the margin destruction caused by agents: inference is currently heavily subsidized. I have serious doubts "on device" inference will ever be reliable and competitive enough to be viable for running high capability agents (will they even be online enough?). What's the real cost of inference? Does Claude Code really cost $200/month at maximum utilization?
It indeed assumes steady state responses from all incumbents and governments while AI agents move at the speed of innovation. Not sure about that.
No comments yet.