On last Thursday, I had the opportunity to go to another team at my company and tell them about our team’s experiences and learnings regarding AI usage and agentic engineering. This appointment happened not coincidently. I had offered my support to other Engineering Team Leads of my company and one of them agreed. I was informed at the beginning that some people of the team are very hesitant to use AI. My goal was not to convert them from skeptics to adopters, I only want to share my motivation and the team’s learnings with them.

I started with the history of how my team started using coding agents and also about the mistakes and learnings we had. During my talk I also emphasized other AI usecases, not only about agentic engineering. I told about our cloud agent, fixing vulnerabilities or also about our product owner writing user stories with codex by having a copy of the backlog as markdown files in a repository.

During the talk, I observed two out of five people to be very reserved, one actually showing the refusal directly with crossed arms. This person was very quite and listened until the end. Then he started with his arguments. He basically said that AI right now is just a hype, and before he starts adopting, he wants to wait until the bubble has burst and the hype is over. He also claimed that the productivity gains are coming from throwing away forty years of engineering practises, and not from AI itself.

He actually caught me on the wrong foot. He argued very rational and fact-based and I honestly didn’t know how to encounter in that situation. Since then, I had some time to think about it.

AI is just a hype

I believe this is true, but my conclusions are different from the ones from my skeptic coworker. Some day, the bubble will burst, but the impact and consequences (for us) are different than we might think.

The technology is here to stay, I’m very certain about that. The losers in the bursting bubble will be the AI frontier model companies such as OpenAI or Anthropic, which are raising and investing money like crazy.

As of this writing (26/4/26), the estimated valuation of OpenAI is $852 billion while Anthropics valuation is “only” $ 380 billion. As far as I know, they both don’t have sustainable business models yet and just live off their investors’ money. When they want see their money back including returns, this will be a bloodbath for them (and others).

Others will do better: Hardware manufacturers, power plant and data center operators, because the demand for AI chips, energy and compute resources will rise.

There are also some factors that could limit or even neutralize the impact of the bursting bubble. The costs per token are getting cheaper and cheaper with each new model and hardware generation. So the invest in hardware is less and less expensive. There are also some companies such as Taalas, that are embedding model into hardware. They are able to produce 17K tokens a second for a Llama 3.1 8B model. Speaking of models, they are also getting more powerful and smaller. Google recently announced the Gemma models, that are producing great results on consumer hardware such as a MacBook Pro.

Throwing away forty years of engineering best practises

His second main argument was that we only achieved performance gains, because we “threw away forty years of engineering best practises” and not because of AI. I think he’s partially true.

I think that agentic engineering doesn’t throw away best practices, but the practises and processes get condensed and can can’t be cleary distinguished. There’s no need for a Planning meeting for the whole team to align the upcoming work. It was only for humans to decide on specific implementation details and to create tasks for parallelization. This is not needed anymore. Given the right context information, the agent designs multiple variants of your feature in no amount of time.

The design and the following implementation can be done by subteams, enabling a whole engineering team of 6+ people to work on 2-3 large features independently.

So the practises and processes still exist, they just have changed to a version where they are unfamiliar to some engineers. Why should you stick to something that doesn’t make sense to you anymore?

---