A new meaning of “agents” in the AI hype era
In the period of the Cold War with Germany and its capitol Berlin being divided into two parts, the term “agent” was used as a synonym for spy. Berlin has been the center of gravity of the secret services, as East and West collided on a very small area.
After the wall went down and Germany got united, the agents moved to other places and it seems they disappeared completely. Today, we see a revival of agents caused by the adoption of AI solutions, and they are all good companions instead of being either friend or foe. All vendors ride the AI wave.
From skepticism to belief
With the short time span of AI in our industry, I’ve been sceptic about the maturity and capabilities of the agents. (I’ve been more like a “Thomas”, the skeptical apostle in the New Testament). But a recent experience turned me into a believer. Agents can do a lot more than just taking over boring tasks that no one wants to do. And there are plenty of them in the world of software engineering.
The live sessions I attended were at the Harness Unscripted conference, showcasing agent-based automations in real workflows.
(Notice: AI is not new, but the current hype and breakthroughs around LLMs and generative automation have brought “agents” back into the spotlight.)
A quantum leap for software engineering
What was so fascinating about the live sessions? Growing up in the time of mainframes and Cobol, agents feel to me like the quantum leap that we dreamt of while taking our first steps writing software with home computers. It was science fiction in those days.
The new reality of self-healing DevOps
Can you imagine CI/CD pipelines automatically and instantaneously healing failures in the code and the pipeline? Or using natural language to embed a UI test for a web application with no need to rewrite it when the UI changes? Just re-rendering? What about using your mother tongue to define security rules and quality gates, embed them in a pipeline, and roll them out to your entire CI/CD stack of pipelines? Finally, updating outdated and vulnerable components in all projects without even knowing where the components are used? And on top, all documented and under your control? I admit that the last example is one of these boring and not very productive tasks that no one wants to do (only as a “shotgun fixing”, i.e. the manager is standing behind the developer with a loaded weapon – that’s the reason why 90% of the components never get updated during the lifespan of an application where it is embedded).
The hidden cost of inefficiency
A study has shown that developers spend between 20-30% of their time on activities that are not really contributing to new features and capabilities of the software. The other way the numbers are even more impressive is that 70-80% of the time your developers work on tasks that do not add visible value to your products by executing tasks that can and should be automated.
Too good to be true? No, real to-day
In case you think “this sounds too good to be true…”, it is reality today! AI initiatives in development teams often focus on accelerating the creation of new source code. But it is equally or even more important to avoid new bottlenecks within your CI/CD environment. It is easy today to generate thousands of lines of code within 15 minutes, but the real work to build, validate, test and release will take a lot longer. We already see that teams can produce more code but experience a lower throughput into production as the CI/CD environment does not scale as needed.
Modernization requires courage
But here’s the caveat – this technology is disruptive to your existing infrastructure! A good analogy is based on a personal experience I had a few years ago. My wife and I bought a 100-year old house and thought about some improvements, as the previous makeover was a while ago. We started to rip out pipes and cables, then continued with the floors … and ended up with blank walls and an open roof. After 2 years of fixing we have an almost new house, easily lasting the next 25 years without a big issue.
Be prepared to face a similar story when modernizing your development infrastructure. It will not pay-off to keep tools alive from the pre-AI era at any cost, although we loved to use them in the past decade. BTW: the next blog is exactly about one of those “champions”…
Get in touch
In the meantime, feel free to contact me at contact@aservo.com to unveil the mystery on how to get your hands on the miracles described above. And remember, there are even good agents out there to make any migration less painful!
Yours sincerely,
Rainer Heinold
