If your agent has to rediscover how to process an invoice every time using computer use or some heavy inference approach it's not going to make it in the enterprise.
The next era is combining intelligence with old school determinism that actually operates in the enterprise.
Reducing inference is actually the goal here.
will be a day very soon when an agent approves 10,000 insurance claims without inspections and someone gets to explain that technically it had permission
because guess what - humans are really bad at consistent clear communication that is actually easy for agents to follow
AI founders: "why are enterprises moving so slowly?"
Enterprise: maybe giving a probabilistic model unlimited freedom to improvise inside our core business systems isn't the greatest idea.
no one wants to buy your shit is just as true in AI as it was in SaaS
the best enterprise AI companies will end up being the ones that figure out how to use the least AI to deliver an outcome
Pre-ai: software ran humans.
Now: humans run software (agents)
no more bs software that is just business logic sitting behind a UI. Agents got that on lock.
First impressions of @OpenAI's new notetaker vs @MonacoGTM:
Not super impressed by OpenAI.
Simple example: Monaco caught the currency - OpenAI said it didn't know
And was way slower to generate the transcript
Does the lab care to make this good or is it a distraction?