| If you are still coding the same way you did two years ago, you might be falling behind without realizing it. Not because you are not capable.
Not because you are not using AI.
But because you are using AI the same way you did when it first appeared. A year ago, we talked how much faster AI could make engineering teams. At that time, speed was the headline. Write a requirement, generate code, move faster than before. The improvement was obvious. After a year of working with AI every day, I’ve realized something more important. Speed is no longer the advantage.
Workflow discipline is. Today I see two types of engineers. Both use AI.
Both move faster than before. One uses AI as an upgraded autocomplete. Prompt, review the diff, merge, move to the next task. The other treats AI as a system component. Before coding, they let the agent clarify the requirement. They separate planning, implementation, and review into explicit phases. They generate tests and extend them with edge cases. They validate critical constraints, the parts of the system where small mistakes create long-term consequences. The difference may look small in the first few months.
Over time, it compounds. One engineer can only focus on one problem at a time because their workflow is still single-threaded. Prompt, wait, review, merge, repeat. The other can scale the number of things they build by increasing the number of agents they can clearly define, coordinate, and control. The difference is not just productivity.
It is leverage. |
I don’t think AI is very good at “plan” compared to someone who is actually experienced with a toolset.
I think the more pertinent “falling behind” aspect for most people is the assumption that the models can’t do complex work. I.e., many people using the AI tools limit what they ask for because they are afraid of getting back bad code they have to fix.
It’s also important to be incredibly specific on what you want wherever possible.
I’ve also tried multiple agent workflows and have found them to generally be tiring and cluttered more than helpful.
Here’s a real life pro tip for you: don’t rush to become amazingly more efficient when a new tool comes along. The only benefactor of that attitude is your employer. I’d rather my employer think that AI is giving them a ~10% boost at best while my workload stays the same. I have a family, I don’t live to work, I work because I have to. Crazy brain-melting shit like multi-agent workflows is antithetical to that.
My last bit of feedback is that this reads too much like a LinkedIn post.
I see now that you’re a director of engineering, and so I now understand the LinkedIn influencer style going on here. Since you’re in a position of leadership, take my advice: don’t expect your ICs to pick up these insane thought leadership workflows that sound amazing on paper but end up causing pain, burnout, and low product quality for the engineers on your team who are actually in the trenches doing the work.
No, you won’t magically get 10x engineers and get to make your CTO happy. Don’t treat AI like a magic pill.
When the Covid-era tech overhiring correction ends, your best employees who have spent 2023-2026 getting squeezed and burned out but haven’t quit due to the job market will be the first to leave when the job market inevitably rebounds. These are the engineers whose dumbass bosses think shove AI down their throats and tell them that they aren’t agentmaxxing sloperator code enough.