Agentic Engineering Has No Elders

Besides many other things, the current shift in software engineering has one big difference from anything before: the missing experts.

Whatever new paradigm was introduced in software engineering in the past, it was always a slow movement. Let's take object oriented programming, which took around 20-25 years to become mainstream. Unit tests took an even longer journey, depending on where you see it rooted. Agile took at least 10-15 years to become mainstream, but the roots of the movements can be traced back to the 1950s.

What any of these have in common is the classical adoption pattern: Early adopters, early majority, … and with that, the experts came. Those were typically part of these two phases and worked in many early projects with the corresponding methodology. They then started to consult other people on that basis and were therefore able to accumulate much more experience and build real expertise. These are the consultants, trainers, and mentors. Kent Beck, Martin Fowler, and many, many more unknown people who built experience and moved it forward - (myself included, for 9 years at Qafoo).

Agentic engineering is totally different. The first real LLM to generate code was seen in 2020. The first people actually doing that in some scale probably around 2023-2024. That's barely 2-3 years of real-scale adoption. And the evolution of LLM code capabilities in this timespan was so extreme, that experience from a project 6 months ago is worthless already today.

For this reason, there cannot be experts on the matter of agentic engineering. Even if you were able to be involved in 3 or 4 projects over the past 6 months, you will not have been able to have built real expertise. And even if you are on the bandwagon for 3 years already, the evolution runs so fast that no human brain can compile down reliable patterns fast enough.

Sure, you can argue that there are Claude, ChatGPT, Gemini and friends now to serve as your experts. But there are 2 issues with this approach: 1. LLMs typically just "recite" what they have "read", even if in widely different forms. If there is no expert literature to read, they will not be able to tell you about it. They probably can consume many hype-cycle articles and condense these for you, but that is far from what any expert from past times could have given you in terms of reliability. 2. If you are not in the early adopter or early majority groups, there is a reason for that. You probably don't have enough trust or not enough resources for experiments. But then, you also don't have these for an LLM consultant to coach you to new endeavors.

Still, FOMO spreads fast in the IT industry nowadays. But not everyone has the money or skills to experiment on their own. So, what should these people do to not miss the train, if there are no experts to guide them? I think there is only the chance to find someone who does NOT claim to be an expert (because there are none), but still has a finger on the pulse and has the consulting & mentoring skills.

What is your opinion on how companies should approach getting into the agentic engineering era?