[Crowberry]

I second this. Most of our customers IT department struggle to look at the responses from their failed API calls. Their systems and organisations are just too big.

As it stands today; just a bit of complexity is all that is required to make AI Agents fail. I expect the gap to narrow over the years of course. But capturing complex business logic and simplifying it will probably be useful and worth paying for a long time into the future.

[agwp]

Also, for many larger companies, access to internal data and systems is only granted to authorized human users and approved applications/agents. Each approval is a separate request.

This means for any "manual" or existing workflow requiring a access to several systems, that requires multiple IT permissions with defined scopes. Even something as simple as a sales rep sending a DocuSign might need:

- CRM access

- DocuSign access

- Possibly access to ERP (if CRM isn't configured to pass signed contract status and value across)

- Possibly access to SharePoint / Power Automate (if finance/legal/someone else has created internal policy or process, e.g. saving a DocuSign PDF to a folder, inputting details for handover to fulfilment or client success, or submitting ticket to finance so invoicing can be set up)

[thorawaytrav]

It is much easier to use an AI API in my bank than to use any other tool. Since the AI is from MS, it's ready to go, whereas other tools require a few months of budgeting, licenses, certs, and so on. Since AI/Azure/AWS is already there and 'certified to use,' it is easier for me to patch something together using this stack than to even ask for open-source software

[j45]

Skills seem to be promising.

I never understood the evolvement around agents, they just appeared to me as Python scripts initially (Crewai 2-3 years ago).

The question is can people see that agents will evolve? Similar to how software evolves to handle the right depth of granularity.