Stop Building AI Agents: Brief and Control Them Safely | Ep 256 | DevReady Podcast

·42 min·2 clips

Gareth compares AI resistance to Luddites smashing cotton mills and even mine sensors with hammers.

1. DevReady Podcast episode 256 focuses on Gareth Ryden’s monthly AI chat about agents, control, and practical use. 2. Gareth Ryden from Friday joins as the recurring guest, and the host frames him as someone tracking what clients are actually doing with AI. 3. The episode asks what people should do with agents now: build them, instruct them, or keep human control in the loop. 4. Gareth says the market has shifted from “what agent are you going to build next?” toward learning how to communicate with prebuilt agents in ChatGPT. 5. He gives a rough split of “85 to 90 percent” of people needing to learn how to interact with agents and only “10 percent” needing to build them. 6. He argues that many LinkedIn claims about building agents are really automations and workflows, not full agents. 7. He walks through a practical example for senior executives: write out a task’s steps, then turn those steps into instructions for an agent. 8. He says people feel more confident once they treat an agent like a person who needs clear instructions. 9. Gareth highlights a control question: when should a user hand over access credentials, and when should they stop at a login screen. 10. He warns that giving an agent full Gmail, Outlook, or calendar access raises murky safety questions. 11. He says ChatGPT Agent opens its own browser, so users should watch the first three runs and notice what sites it visits and how it behaves. 12. He notes that the browser shows little notation about why the agent clicks, which makes observation part of the learning process. 13. He says scraping websites with an agent can be possible, but one prototype reached about $3 per page and 127 hours of continuous use. 14. He uses that cost example to argue that repeatable workflows often belong in pipelines, not in a single open-ended agent. 15. He compares AI resistance to Luddites destroying cotton mills and to mine-site sensors being smashed with hammers. 16. He says this pattern shows how people may try to protect jobs, but the technology and the underlying demand keep moving. 17. The conversation also covers people who fear AI taking their jobs and then refuse to use AI at all. 18. Gareth’s tone is practical and consultative, with examples from workshops, coaching sessions, finance, legal work, and enterprise software. 19. People who manage teams, workflows, or AI adoption in businesses would get the most from this episode. 20. People looking for product news only or a hype-driven AI recap may skip it.

As heard by us

A practical conversation about prebuilt agents, deployment risk, and the human oversight AI still needs.

Anthony and Gareth keep the discussion on practical ground. They look at prebuilt agents in ChatGPT instead of building them from scratch, and they do not dodge the tradeoff: quick usefulness on one side, prompt injection and hacking risk on the other when an agent gets released…

Read the full review in PlayNext →

Why you'd press play

If you need a reality check on ChatGPT agents, start here.

Read the full recommendation in PlayNext →

Listen to the show on

Apple Podcasts Spotify

Play clips from this episode More from DevReady Podcast