Something I’ve been mulling over for a while now relates to where our industry is going with all of the change we’ve been seeing over the past year or so - a move away from hand writing code to instead being expected to orchestrate a fleet of agents to do the work for us. While the jury is still out on the actual outcomes this change will drive, and if it truly becomes a renaissance in software deliverability, it doesn’t mean we don’t need to figure out each of our own places in that future.
I’ve been in the industry for 15 years now professionally and primarily leading engineers (and teams of engineers) for 10 of that. The majority of my time hasn’t been actually writing the code but I’ve been talking with a lot of my peers who’ve been doing this for quite some time in various other roles - the common theme that I’m picking up on is the majority of them find no value in telling an agent what to do and are looking for ways out.
I would imagine the majority of us that got into this line of work got in because we enjoyed the craft behind it. We loved diving deeper into the inner workings of what was actually happening or how to best leverage the limited resources available to achieve something novel or performative. Finally cracking the final bug, design choice or API call that resulted in the aha moment felt like magic.
Does that still happen when you tell Claude to plan out a solution with you? Is the magic the same if you cracking the nut has moved to cracking nut for the prompt that gets the agent to crack the nut for you? Do we still have the deep sense of ownership of our service in production if we primarily contributed to its initial planning but can’t answer every intimate detail about its development when a bug arises?
This line of thinking is really tough, and I think that it’s going to push a lot of people away from this line of work.
What does it mean for managers?
Meta recently announced it’s testing out an ultra-flat organization design given the advancements of AI within their organization: a 50:1 ratio of engineers to managers. The idea here is for the managers to primarily communicate and oversee their reports via automated, LLM driven, workflows to manage performance and ensure business continuity.
While it’s still early for them, and we do need to note that Meta is incentivized to drive this type of thinking given its own product launches in the AI space, I personally am not excited for a work like that to exist.
Managing, leading, people is an inherently human thing to do and one of the reasons I still do this work to this day. Software Engineering is a people problem, not a technology driven one. Getting smart people to work well with one another so that the sum of their outcomes are greater than they would be individually is challenging and always changing. I hope I never get to the point where the expectation is that I have to talk to someone via a several-level generated report from some LLMs.
Jury’s still out on how the rest of this is going to go…
