This blog's namesake drives my writing process.

I first come across a noteworthy book or article, something that triggers a thought or reaction. Then I create a new doc, jot down a few thoughts, and let it sit for some time before fleshing out the idea more holistically. Slow introspection, if you will.

Lately, I've been collecting more and more scattered thoughts on engineering management that I can't seem to render into anything publishable. We've known that our discipline is in flux—careful consideration is losing out to parabolic progress.

COVID acceleration

Our industry rode this massive wave and a corresponding trough in the aftermath of COVID. Low interest rates, plus sheltering at home, supercharged all things digital: e-commerce, remote tooling, and broadly "things that kept people entertained while isolated" outperformed. Tech companies responded logically, by ramping up hiring, scaling out teams and products, and leaning into the pandemic accelerant. I recall 2021 as a year of excess, of meme finances and crypto and IPOs and SPACs.

Of course, all manias end; all the pandemic stimulus and disruption showed up a year later as global inflation, which in turn led to raising interest rates and the official end of the Zero Interest Rate Policy (ZIRP) era. Most of tech, over the course of 18 months of interest rate hikes and softened demand, reverted back to pre-pandemic norms—though it didn't feel like a reversion as much as a collapse in the moment.

Management in Flux

During this time, I noticed that the engineering management communities I participate in, the ones that advocated for mentorship and best practices, were already shifting.

This was our new reality of smaller budgets and reduced headcount. The cliché has always been to "do more with less," but now it was no longer a pithy slogan for efficiency—CFOs and CEOs were actually scrutinizing engineering organization run rates. We started tracking en masse layoffs, again.

When businesses cannot fully sustain their accumulated employee base from the go-go years post-pandemic, our management instincts around hiring, onboarding, career development, etc. became distant luxuries.

Other engineering leaders had noticed this shift earlier. The very idea of good management seemed like a fad that had run its course. And the job was no longer as much fun.

AI Destabilization

If the pandemic and its repercussions had been the only black swan event of the decade, we could have followed a crest-to-trough wave that is so familiar to the tech industry.

But for whatever reason, San Francisco and the Bay Area have been the epicenter of boom-bust cycles, and the AI wave started rising right on the back of the COVID retrenchment. By 2026, we've transformed our industry again.

AI—in both its coding and agentic forms—reset expectations around efficiency. Software engineers are expected to produce multiplicatively more output, by changing their everyday workflows from direct hands-on development to directing, reviewing, and approving AI-generated code. Github is seeing exponential growth in developer activity, and rather than just sharing code, engineers are showing off their agents.md files.

The job of the software engineer now resembles the archetype of the tech lead/manager (TLM). In turn, their immediate people managers cascades, to something close to that of a "manager of managers" or an engineering director.

This makes some intuitive sense. As each engineer's scope increases with AI assistance, the overall scope of the team correspondingly widens. Along with layoffs and executive efforts to flatten organizations, some managers now operate with the same breadth of responsibility as engineering directors, pre-agentic AI.

Hollowed-out Middle Management

Take the pressure to its conclusion. The net effect of waterfalling roles in squeezing out the middle management layers, at least in the short run.

Engineering directors, who have risen from line management to be able to run multiple teams and domains are suddenly finding that skillset obsolete. Previously, the atomic unit of a worker is a software engineer, and the tools to engagement were skip-level 1:1s, team get-togethers, staff meetings and such. If the atomic unit becomes an AI agent, such formalities have less impact; they may be outright counterproductive.

A related second-order effect is that team size is less useful as a proxy for scope and competence. And in the age of tokenmaxxing, we shouldn't jump and replace it with its AI equivalent either.

We should be concerned about the long-term impact as well. Just as there's a clear progression on the software engineering career ladder from entry-level → senior → staff → principal, the management ladder goes something like supervisor → manager → director → VP → CTO. Each level builds upon the previous, and it's apparent when managers who excel at one level become overwhelmed when asked to execute 2+ levels above. We are destroying the management talent pipeline on the altar of efficiency.

What Even is a Best Practice Anymore

In my last role, one of our VPs spent time with DX to measure developer productivity. He was able to quantify engineering output, at a team level as well as at the organizational level. We eventually added and cross-referenced data on AI usage, again by team and by org.

But despite clear warnings that productivity metrics shouldn't be relied upon for performance evaluations, we succumbed to the pressure from the CEO and the board and made personnel changes based in large part on this data. Yet, despite my reservations, this has since become an industry-wide trend; the talent budget shortfalls a few years ago have suddenly become AI budget excesses. I can't blame engineers—and their managers—for responding aggressively to overt incentives around using as much AI as they can.

Then again, this pattern is itself only a few months old. Our actions around DX were only a year old as I write this, and I have no idea what will materialize as proven wisdom 6 months from now. The frameworks that we've built in the 2000s and 2010s no longer apply with the same level of strength and conviction, and AI is moving too quickly for us to find a new normal right now.

Which means I keep on reviewing, and deleting, the notes in my engineering management drafts folder. And wait to see what tomorrow brings.