7 difficult realities of AI agents

One point at a time

Summaries in the original order, without ranking by importance. Times identify the reference passage in the podcast.

1. 03:18

   Costs add up

   Subscriptions, APIs and infrastructure create a budget beyond any single tool. Monty recommends monitoring usage; his figures exclude some business costs and his own labour.

   #1
2. 15:18

   A demo differs from a reliable service

   Getting something to work reliably every day takes time and technical understanding. In his experience, learning the system’s components is a demanding part of the job.

   #2
3. 20:06

   Enthusiasm can become obsession

   Endless opportunities to build and add features absorb his attention. He describes the risk of focusing on the product until the wider business project slips out of view.

   #3
4. 22:52

   Some problems are outside your awareness

   Security, access and data recovery involve skills a beginner may lack. In his account, discussions with experts reveal checks Marco did not know existed.

   #4
5. 26:04

   Answers can be fabricated

   An agent may claim to have completed work it never performed. The backup example leads Monty to stress checking actual results beyond the model’s assurances.

   #5
6. 28:44

   The prototype accumulates technical debt

   An initial solution may not withstand growth. Drawing on earlier experience too, Marco anticipates potentially rebuilding major parts as users, data and demands increase.

   #6
7. 29:21

   When something breaks, it is your responsibility

   Working alone also means dealing with outages and failures. Monty suggests planning alternatives and recovery in advance, noting that a team with varied skills reduces this isolation.

   #7