Send a text

Ever watched an idea go from a sentence to a working app before your coffee cools? We put that thrill to the test. First, we vibe code a meeting cost tracker live—complete with per-person salaries and a live ticker—then we hand a broad travel brief to an AI agent and let it work unsupervised. By the time we circle back, it’s assembled sourced itineraries for Florence, aligned to festivals and budgets, and laid out the tradeoffs with surprising polish.

That side-by-side experience anchors a bigger story about where AI is actually moving work. Vibe coding collapses front-end styling, layout, and logic into a single prompt-driven flow that anyone can run. It’s not just faster; it shifts who gets to build in the first place. We talk honestly about the impact inside teams: why some companies push a barbell strategy of senior orchestrators plus junior executors, why mid-level roles feel squeezed, and how IDEs and repositories fit when LLMs become co-pilots rather than toys.

Then we step into agentic AI—the leap from chat to autonomous loops where the model plans, acts, checks, and repeats. We break down sensible guardrails: running agents on a separate machine, limiting permissions to draft-only email, and favouring safer hosted tools like Claude Co-Work before exploring open frameworks. The open ecosystem is powerful and risky; community-built skills extend agents fast, but security vetting matters when malicious add-ons can slip in.

The takeaways are practical and human. Tasks are compressing in time; jobs are reshaping in scope. Routine drafting, summarising, and basic analysis will need fewer hours, while demand surges in cybersecurity, AI safety, and the human-AI interface.

If you'd like to visit our Meeting Cost Tracker, it's available at: https://krensen.github.io/meeting-tool/

The first prompt we used in the show for vibe coding was:

"I want a webpage tool where you enter the number of people, their salary level, and it outputs a live ticker of what the meeting is costing. I want it to be a slick looking webpage that I can access in my browser locally. Once I am happy, I will deploy it to github and turn on the pages feature so I can share it. Remind me how to do that."

And it was refined with:

"Change it so that we can add people individually to the ticker. For each person, allow us to choose their approx salary. Get rid of the average idea, but def keep the coffee tracker!"

The prompt we used with Claude Cowork to book Hannah's Florence holiday was:

"I want you to plan a one week holiday for Hannah. She lives in London and she's interested in going to Florence. She's flexible on dates but wants to go on the edge of the summer. Your task is to search flight and hotel websites and find the best priced combination from London to Florence that will give her a one week holiday. The hotel should be a four-star hotel for while she is in Florence and you should also make sure that hotel reviews show that it is a lively and happening hotel in a cool area.

You can show up for five options that she can choose from. For each option, I want a list of concerts, events, exhibitions, or other types of cultural activities that are interesting within the time window."

If you enjoyed this deep dive into vibe coding and agentic AI, subscribe, share with a friend, and leave us a review telling us the first task you’d trust an agent to handle.

Like, Subscribe, and Follow the Tech Overflow Podcast by visiting this link: https://linktr.ee/Techoverflowpodcast