Twenty-seven years ago I nearly put six people out of work with an Excel spreadsheet.
Every AI conversation I have now takes me back to it.
I was 22, a fresh graduate, three months into an internship at Marconi’s underwater weapons division in the Spearfish trials analysis team. (My future pacifism started here - turns out spending your summer analysing torpedo data doesn’t sit well long-term.) The team I joined had a job about as glamorous as it sounds: read printouts of hex data from torpedo test runs, decode them by hand using highlighter pens, type the results into Excel, generate charts, build a Word document, and finally - finally - do the actual analysis.
Steps one through six took a week. Step seven, the bit that needed a brain, took an afternoon.
So I did what any 22-year-old engineer with no political instincts would do. I automated it.
STAR WARS
Four weeks of VBA later, I had a single Excel add-in that pulled trial data straight off the mainframe, decoded the hex, charted the torpedo’s path, and spat out a finished Word document ready for analysis. Total runtime: about ten minutes.
I gave it a name: Spearfish Trials Analysis Report Writers Automated Reporting Suite. STAR WARS for short. (Episode 1 had just come out. Cut me some slack.)
I was very pleased with myself.
The reception
Everyone hated it.
One of the senior report writers challenged me to produce a better report than hers, from the same trial data. I had mine in ten minutes. She had hers in four days, and it contained several mistakes mine didn’t. She did not take that well.
After I left, the spreadsheet was deleted. Everyone forgot it existed.
It took me years to understand why.
STAR WARS could have replaced six of the seven people on that team. Six people with families, mortgages, kids in school, and a quiet path to a Ministry of Defence pension. The senior report writer who hated my four-day Word document wasn’t defending the quality of her work - she was defending her ability to feed her family. I just couldn’t see it at the time.
I thought I was being a hero. I was being a threat.
The same fear, dressed up differently
The conversation about AI and jobs in 2026 sounds new but it isn’t. The fear I saw at Marconi in 1999 is the same fear running through every meeting room and Slack channel now - it just looks more sophisticated.
Watch for the modern signs:
- The senior engineer who won’t turn on Copilot, who has reasons (privacy, code quality, taste), and one unspoken reason that’s louder than the rest.
- The product manager who suddenly stops documenting her workflow because the company is “piloting an AI process improvement tool” and she’s not stupid.
- The team that, when surveyed about AI usage, all reports they “use it a little” - which means either everyone is dramatically under-reporting or dramatically over-reporting, depending on what their manager wants to hear that month.
- The artisanal inefficiency - someone keeping a manual step in a process they could automate in an afternoon, and quietly hoping nobody notices.
This is the 2026 version of the Marconi report writer pretending her four-day Word document was better than my ten-minute Excel macro. Same instinct, different decade.
And just like in 1999, the people doing it aren’t being dishonest because they’re bad people. They’re being dishonest because they’re afraid, and the system has given them every reason to be.
The intern problem
There’s something specifically 2026 about this that I didn’t see in 1999. At Marconi, I was the intern who could have eliminated six jobs. Now, the intern is what’s getting eliminated.
The career ladder in a lot of disciplines has lost its bottom rung. AI hasn’t taken the job - it’s taken the entry-level version of the job that everyone used to use as their training ground. Junior dev hiring is down. Junior copywriting roles are evaporating. Paralegal entry-level work is being absorbed at speed. The senior people are mostly fine for now. The seniors of 2036, the ones who would have started their careers in 2026, are quietly being squeezed out before they ever got to start.
Nobody talks about this on LinkedIn because the people most affected don’t have LinkedIn followings yet. They’re not in the conversation because they haven’t been hired into it.
What I think we actually owe each other
I’m not against AI. I work in AI. I would absolutely automate those torpedo reports again - they were a waste of human time and brainpower, and pretending otherwise wasn’t kindness, it was just delayed pain.
But the part we can fix is the dishonesty.
- Stop telling your team AI is “augmenting” them when you’re using it to thin them. They can read a headcount plan.
- Stop running internal “innovation workshops” where you ask people to describe their workflows for automation while calling it empowerment.
- If you’re cutting jobs because AI does them now, say so. Cut honestly, support people through it, and don’t dress it up as a productivity initiative.
- Better yet, don’t cut. The task is automated. The person isn’t. The Marconi report writers had years of domain knowledge about how torpedoes behaved, what patterns in the data meant something was wrong, what mistakes the engineering teams kept making. STAR WARS replaced the typing. It couldn’t replace any of that. If you’ve automated the routine work, you now have a team with deep domain expertise and suddenly a lot more time. That’s a growth opportunity, not a headcount line. Redeploy people into the strategic work nobody ever had bandwidth for. Let them shape the next product, train the AI properly, catch the failure modes the model can’t see, mentor the juniors that do get hired. Replacing experienced people with AI is the lazy version. Growing the business with the same people doing more valuable work is the harder one - and the one that doesn’t leave you rebuilding institutional knowledge from scratch three years from now when you realise the AI never knew why things were done that way.
- If you’re a leader and you’re not thinking about what happens to the bottom rung of your industry’s career ladder, you’re storing up a problem that you personally won’t feel for ten years, but somebody will.
There are real conversations to be had. UBI pilots have actual data behind them now. Taxation of large tech companies is back on several governments’ agendas. Apprenticeship models are being rebuilt from scratch in some industries. These are good conversations and I’d rather have them now than after the dust has settled and the damage has been done.
But none of them work if the people running the change aren’t willing to be honest about what they’re doing.
The fear at Marconi was rational. The fear now is rational. The dishonesty about it is the part we can actually choose to fix.
This is an update to a post I first wrote in 2023: Avoiding automation to keep your job. The Marconi story is the same. Everything else has moved faster than I’d expected.