Deep Agents
So today, a lesson was learned
After listening to the 'talk python to me' podcast I became very interested in Deep Agents from Langchain, and with the wide eyed idiocy of a curious mind that has absolutely no idea what he is doing, set about building a little agent that could look at a headline, check its validity, then report back writing a file. All cool stuff.
It went well! A few bugs (Im just learning python) and i have already built a shallow, dumber agent called paperboy that gets my morning articles and delivers them to my mailbox.
So after the initial success of my new agent, I thought "I wonder what he can really do? I wonder if he can summarize an article and THEN write a report". Well, of course he can, if the code is written by a competent coder with experience, knowledge and skill.
But let an idiot loose with another ai as backup writing the code that he doesn't understand, and you have yourself a spiral of disaster, or in the agents case, a never ending loop of token spending that wouldn't stop.
"Its taking a while" I happily plugged into my ai partner in crime, "Yep" she said, "its probably running queries, I would..." Brilliant ok so its working.
Luckily I got bored waiting and thought "Hmm,... something is amiss here" and dropped in to Anthropic to check the balance and there we had in one session over 7.5 million tokens... oh dear.
Then my email buzzed and a few polite emails in succession told me that i had busted through the three warnings that I had set up in succession, and THEN said i had been cut off and owed Claude some cash...
Thankfully I pre paid my api and didn't have it on auto refresh and had set a monthly limit, holy crap can you imagine?
So lesson learned. Don't mess with the big boy stuff until you know what you are doing.
...So I just found out about Open Claw...