Opinions.
The tracks are organised by your stage. These aren't. This is the case for a specific, arguable position, held long enough to write down, not softened to survive a comment section.
Where a lesson tells you what to do at twenty engineers, an opinion tells you why the ground moved under all of it. Disagree with it. That's what it's for.
Intent architecture
For thirty years, the gap between what an organisation actually wants and what it writes down got closed by a person, every time, without anyone noticing it was happening. The moment a system is doing the reading instead of a person, that gap stops closing itself.
Includes a toolAdapted from After the PromptThe prompt tax
Every recurring prompt looks like a productivity win because you're measuring the wrong side of the ledger. The real cost isn't building the workflow. It's operating it, every week, for as long as the team runs it.
Includes a toolThe monolith default
Ninety percent of production software would ship faster and break less as one deployable than as the distributed system most teams reach for instead. Not for a first year before graduating to services. As the actual architecture, for as long as the product exists.
The Kubernetes tax
Kubernetes solves real problems that almost no company under thirty engineers actually has yet. The bill isn't the weekend it takes to set up. It's every subsequent debugging session running through a layer nobody on a small team fully understands.
The agentic prerequisite
Every conversation about getting ready for agentic development is about the tool. The variable that actually decides whether it goes well is whether your test suite still tells the truth, and agents don't create that discipline. They just find out immediately whether you have it.
Nobody has to approve the agent's merge
The mandatory-human-approval rule most orgs wrote this year is theatre. What gates a change to main was never the signature at the bottom of the pull request, it was the machinery underneath it, and the only thing agents changed is that the difference is now visible.
Adoption isn't a result
Percentage of engineers active weekly, percentage of merged code AI-generated, prompts per head. Every one of those is a utilisation number, and optimising utilisation is how organisations made themselves slower for decades before any of this arrived.
Your AI bill is a design defect
Token spend scales with decisions inside your workflows, not with how many people hold a licence. That makes it an engineering problem arriving in finance's clothing, and every control finance reaches for first is aimed at the wrong variable.
The invention floor just dropped
The scarce skill was never having the idea. It was the distance between having one and putting it in front of a stranger, and for a huge class of ideas that distance now runs to a weekend instead of a headcount. Function doesn't matter. Finishing does.
Never rewrite was a budget rule
The rule held for twenty-five years because rewriting cost human-years. That bill has collapsed and the other one hasn't moved, which means the thing that used to stop bad rewrites has quietly stopped stopping them.
The substitution trap
Buy the assistant, measure the hours saved, report fifteen to twenty percent, collect the green check. The ceiling on that story is set the day you define success as doing the existing work faster, and the maximum upside is a fraction of your current labour cost.
The org chart is the bottleneck
When electricity replaced steam, factories bolted the new motor into the old engine's spot and kept the belts. It took about thirty years to redesign the floor. We're in that gap now, and reorganising around AI-era workflows matters more than which tools or models a team picks.
The connective tissue
The "AI taking jobs" framing misses the thing actually worth worrying about. Teams are how people make sense of work, and when the team dissolves into individual operators with agents, we get more efficient and lose something nobody is measuring. I don't know how to prevent that.
The gatekeeping reflex
When the pace of possible change outruns what a team can absorb, organisations build friction: new approval workflows, weekly review boards, an AI governance committee adding three days to a two-hour build. The gates don't improve quality. They reduce anxiety, and that is a different job.
Consensus breaks before the team does
Consensus is a serial process running on top of a parallelised implementation layer. It can't keep pace with a cycle that goes from spec to production in ninety minutes, and no framework fixes that. Either one person decides or nobody does.
Leaders must stay hands on
For twenty years the advice was that the best leaders don't write code, they build teams that write code. That advice is wrong now, not because you need to prove you can still code, but because every other decision you make sits on technical judgment you can only keep calibrated by touching the work.
Elite is now table stakes
The metrics that meant world class for a decade are the floor now, and the processes that earned them have turned into friction. The hardest part isn't the tooling. It's that the organisations best at the old thing have the most to unlearn.
Code freezes are a confession
A code freeze is a public statement that you don't trust your own safety mechanisms. It gets described as prudence before a busy period, and what it actually does is stockpile every change until the system is under the most load, then release them all at once.
Speed is an outcome, not a value
Every fast company I've seen became fast the same way. They removed the things that made work slow. Not one of them got there by deciding speed was a value, and the ones that tried mostly ended up with the word in the handbook and the old pace on the ground.
Service as software
CI/CD is still plumbing. You write the YAML, configure the triggers, and the pipeline does what you told it. The version worth paying for knows your codebase, decides what to validate, and charges you for successful deployments rather than pipeline minutes.
Your vulnerabilities are yours
Every senior engineer should have an agent hunting for security holes in the systems they own, all the time. Not once a year before the audit, not when the security team gets to it. The excuses ran out when the cost of looking dropped to nearly nothing, and so did the idea that finding the holes was somebody else's job.