“`html
Key Takeaways
- Vibe coding and agentic engineering are starting to blur in Simon’s work.
- Simon is comfortable using agents like Claude Code without reviewing their code, due to the tools’ reliability.
- The normalization of deviance applies: trusting a model without reviewing its output can lead to future issues if not handled carefully.
- Simon values real-world usage over automated tests and documentation for software evaluations.
- There are new challenges in evaluating software quality, as development cycles have shifted with the speed at which code is produced.
- Simon remains confident about his career because these tools amplify existing experience; producing software is still a difficult task that AI can assist but not replace.
Vibe coding and agentic engineering are starting to overlap
When Joseph brought up the distinction between vibe coding (where you’re not looking at the code) and agentic engineering, Simon had a sudden realization that they were not as distinct for him as he initially thought.
“Weirdly though, those things have started to blur for me already, which is quite upsetting.”
For vibe coding: “not looking at the code at all. You might not even know how to program. Asking for a thing and getting it done if it works.” This contrasts with agentic engineering where Simon considers himself a professional software engineer using these tools.
The key difference is that in vibe coding, responsibility lies with the person asking for something; in agentic engineering, it shifts to the tool or model.
The new challenge of evaluating software
Previously, if a project had a hundred commits and good documentation, you could trust that care was taken. Now, creating similar projects in half an hour is seen as equally valuable.
Simon values the actual usage more than just having tests and documentation; he wants to see how something works in real-world scenarios before trusting it heavily.
The bottlenecks have shifted
Previously, a day of writing code was the standard. Now that we can produce 200 lines per day easily, what else breaks in our development process?
Jenny Wen from Anthropic pointed out that design processes are often designed for projects taking months to build. With faster production cycles, these risks decrease.
Why I’m still not afraid for my career
These tools amplify existing experience; if you know what you’re doing, you can run much faster. However, producing software is still a ferociously difficult task.
Simon remains confident in his role as a software engineer because these AI tools are just amplifiers of existing skills and knowledge.
The impact on SaaS providers
Simon now wants to use solutions that have been proven successful by at least two other enterprises before he takes a risk. This reflects his trust in existing systems rather than new, untested ones.
“`
Stay ahead of AI. Get the most important stories delivered to your inbox — no spam, no noise.
