Agentic AI is interesting because it has the opportunity to create competitive advantage that other companies cannot replicate. And that’s the promise. Yeah, you can pay someone to build a custom agent, but it’s $50k+ plus maintenance for one agent - and it just doesn’t scale. It’s way cheaper to train up your team who can build many agents.
Companies that delay or avoid training their teams are in for a bad surprise when a new company built around the technology enters the market. It will be one an existential problem that will result in job loss (think Amazon vs brick and mortar).
My general recommendation for workers is to get trained up on agent building - then go get a job at a company investing in their people, because those are the places most likely to survive.
Of course all this will collapse when the bills come due and people have to pay the actual cost of LLM “tokens”. As users of Microsoft’s Copilot found out not long ago.
Agentic AI is interesting because it has the opportunity to create competitive advantage that other companies cannot replicate. And that’s the promise. Yeah, you can pay someone to build a custom agent, but it’s $50k+ plus maintenance for one agent - and it just doesn’t scale. It’s way cheaper to train up your team who can build many agents.
Companies that delay or avoid training their teams are in for a bad surprise when a new company built around the technology enters the market. It will be one an existential problem that will result in job loss (think Amazon vs brick and mortar).
My general recommendation for workers is to get trained up on agent building - then go get a job at a company investing in their people, because those are the places most likely to survive.
Of course all this will collapse when the bills come due and people have to pay the actual cost of LLM “tokens”. As users of Microsoft’s Copilot found out not long ago.
I mean, you can put caps on token usage at the user level. Also, you can test while building so you have an estimated cost per run.
When you look at the costs involved, putting caps on means you run out of tokens in your first morning of use per month.
The sticker shock for the Copilot users was a couple orders of magnitude.
We capped everyone at $50/month. You can see how many tokens you use per task my typing /cost. Then you can manage it.
I’m not sure I can justify those other people’s poor/stupid behavior.