1 pointby Socrates062 hours ago2 comments
  • gnatan hour ago
    Fast is probably going to get in the way of Good, but here are some things to think about.

    1. In many businesses, different users of the agent should have different read/write permissions to data.

    2. Log everything. There will be problems. You'll need to debug, attribute, defend yourself against accusations that the cockup was your fault, etc.

    3. If you're just providing read-only then the more you can enforce that at all layers of the stack (e.g., read-only user in the database).

    4. Don't give the agent direct SQL access to the database. Write functions in your favourite programming language, expose them as tools to the LLM.

    5. Don't make your LLM do maths in its head, they're crap at it. I had some success giving the agent tools that fetch data from the database (where I control the SQL that runs) into a SQLite in agent memory, then the agent can run SQL against that database. Gives you a chance to put the data into an obvious, easy to query, format rather than whatever arcane historic chaotic state your main schema has evolved to).

    • Eapz_06an hour ago
      Thank you so much... One doubt I have is when we expose tools instead of this raw sql, how do ensure that agents are using the correct data like right source and not stale.

      Do we have to enforce some checks manually or is it not worth to worry(now it's read only right).

      And one more doubt, what's the reason u introduced a lite layer in between?

  • andsoitisan hour ago
    What are the goals of this AI agent?
    • Eapz_06an hour ago
      So it's nothing much ..just basically answering some questions from their data pool, need to check inventory. Still il the read only version.
      • andsoitisan hour ago
        Who are the users of this AI agent?
        • Eapz_06an hour ago
          Mostly internal employees itself.