20 years of IT: soft skills, Big Tech, and the hype cycles I watched die

I entered this industry when a framework was a personality. Struts, then Rails, then Node, then whatever was going to kill the last one. I have been adjacent to Big Tech hiring loops and I have been inside small companies that tried to copy those loops. I have watched SOA, then microservices, then “maybe a modular monolith” as if we invented folding clothes. Twenty years is long enough to see the same slide with a new logo. It is also long enough to notice what did not die.

This is not a memoir of titles. It is the short list of soft skills I would still hire for, the Big Tech habits I would steal, and the hype cycles I personally watched go quiet — plus the one or two that quietly became plumbing.

The soft skills that outlived every stack

Writing a decision so someone else can disagree with it. I used to think this was “documentation.” It is not a wiki. It is a paragraph that says what we will not do. The people who lasted were the ones who could put that paragraph in Slack before they put a PR up. The people who burned out often wrote excellent code into a vacuum and then were surprised when sales promised the opposite.

Estimating as a conversation, not as a number you will be beaten with. I got better when I started saying “this is a week if the vendor API is what they wrote, and a month if it is what they shipped.” That sentence is a soft skill. It is also an engineering skill. I do not separate them anymore.

Sitting with a user or a support ticket until the bug has a name the user would recognize. I have watched brilliant backend people refuse this and then optimize the wrong query. Empathy here is not a poster. It is a debugging technique.

Saying “I do not know” early enough that someone can still change the plan. Twenty years ago I thought that was weakness. I now treat the opposite — confident invention — as the expensive failure. Juniors get this backwards because we reward fluency in interviews. I have tried to reward the flinch in 1:1s instead.

The skill I still cannot automate: telling a founder or a VP that the date and the scope cannot both be true, without becoming the person who is “not a team player.” I have failed this. I have also done it well once or twice. It is the same muscle as a good design review. The audience is just more expensive.

Big Tech, as I actually experienced it

I have not spent twenty years on a campus with a climbing wall. I have spent years hiring people who had, and years competing with companies that did. What Big Tech actually bought those people was not a secret algorithm. It was: a design review that could say no, a pager rotation with a follow-up, and a promotion packet that forced you to write down impact. Those are stealable. A 12-person company will not steal the six-layer interview or the skip-level theater. I have watched small companies import the theater and skip the review. That is how you get a “staff” title and a still-on-fire on-call.

What I would steal: written design for anything that touches money or deletes data. A blameless-enough postmortem that produces a change, not a novel. Interview loops that include a real artifact from your domain, not only a graph puzzle. What I would not steal: calibration as a blood sport, transfer as the only growth path, and a process that needs a process team to explain the process.

Compensation theater is part of the twenty years. Equity stories, refreshers, the $1M headline that is usually a few years of stock in a hot window. I tell juniors to read the vest and the refresh, not the blog. I will write more about that math another time if I have to. The soft skill is not worshiping the logo. The soft skill is knowing what you are trading for the logo: scope, hours, and whether you will still own a user outcome.

Hype cycles I watched die (or become boring)

SOA and the ESB. Every company needed a bus. Then the bus was the product. Then we called the same idea microservices and used HTTP. The dead part was the vendor bus as strategy. The living part is “do not share a database if you cannot share an on-call.”

NoSQL will replace Postgres. I liked Redis. I liked S3. I liked a document store for a document. I watched teams put invoices in a store that could not join and then invent joins in the app. Postgres absorbed JSON and most of the rebellion. The cycle died as a religion. The tools remained as tools.

The SPA as the only grown-up UI. I shipped those. I also shipped the hangover: a 200KB hydrate to toggle a checkbox. In 2026 the hangover is named. HTML and islands came back. I wrote about that frontend bet already. The hype died. The client app did not, for the apps that are actually apps.

Blockchain for the supply chain / the login / the newspaper. I sat through the decks. A few settlement systems remain. The rest is a punchline. The soft skill was leaving the room without calling your counterpart a fool. Sometimes they were your CEO.

AI will replace the junior in 18 months. I have used the assistants. They write a first draft. They weaken a test if you let them. The junior who can tell is more valuable, not less. The cycle is not dead. It is loud. I treat it like the last cycle: use the tool, keep the review, do not staff a company on a demo.

Agile as a certified personality. The demo lasted. The two-week hostage note is what I would bury. I already said so in the Scrum leftover. Twenty years taught me that methodologies die when they become a KPI. Feedback does not die.

What lasted when the logo changed

SQL. HTTP. A test you can run in CI. Logs you can grep. A backup you have restored. Version control used like a seatbelt. Those are not hype. They are plumbing. I have watched people skip them for a platform and then rebuild them inside the platform.

Unix as a habit, even on Windows. I got better when I could follow a process, a port, and a file. Cloud did not repeal that. It hid it behind a console. The people who lasted could still find the process.

Types, eventually, in more languages. Not Haskell as a job — I would still learn it as a teacher — but the idea that illegal states should be hard. That idea survived. The language wars did not crown a king. They crowned a habit.

Owning a production path. Twenty years ago we called it “being senior.” We still do. The title inflation changed. The path did not: can you take a page at 2 a.m. and leave the system better at 10 a.m.?

The hiring loop I grew to distrust

I have run whiteboard loops and I have run take-homes. Neither is the job. The job is a week of imperfect tickets and a Slack thread that needs a decision. The best signal I found was a conversation about a system the candidate already shipped: what they would undo, what paged, what they would not do again. Twenty years of hype made us copy FAANG puzzles because they looked objective. They were objective about puzzles. I still use a small exercise so I am not guessing at syntax. I will not pretend a Hard problem forecasts who will write the paragraph that says no.

Remote work did not invent async, but it punished the people whose only soft skill was being in the room. The people who lasted wrote things down. That is the through-line again. I would hire for the written trail even if we all sat in one office. The office was never the skill.

What I would tell myself in year two

Stop collecting frameworks. Collect incidents you understood. Stop waiting for a Big Tech offer to feel legitimate. Steal the written review and the restore drill. Learn to write the paragraph that says no. Learn SQL until you are dangerous. Learn to talk to the person who takes the customer call. Do not confuse a hype cycle with a career. The cycle wants a talk. The career wants a system that still works when the talk is over.

I am still susceptible. I have wanted the new thing. The twenty-year trick is not cynicism. It is a delay: I let a tool survive contact with a backup and a junior before I make it my personality. Most hype dies in that delay. The rest becomes the next decade’s plumbing. I can live with that ratio. I could not live with believing every slide.