Insights / Visibility & Authority
What earlier technology shifts teach us about using AI well
Spreadsheets and desktop publishing made some work cheap and some judgment more valuable. AI is doing the same. Diagnose the work before you automate it.
Every few years a tool arrives that makes some piece of work dramatically cheaper. The first reaction is usually the same: this changes everything.
It does change things. Just not always the things people expect.
AI is the current version. It can draft ten ad headlines in seconds, summarize a report, transcribe a meeting and suggest a page structure. That’s useful. But the businesses that get the most from it will be the ones that ask a better question first.
Not “what can AI do?” but “which of our work should change, and where does judgment still matter?”
Two earlier shifts help answer that.
Spreadsheets made the arithmetic cheap
VisiCalc, the first spreadsheet program for personal computers, came out in 1979 on the Apple II. Lotus 1-2-3 followed in 1983, then Excel. Before them, recalculating a financial plan meant redoing columns of figures by hand. Change one assumption and you started again.
Spreadsheets made recalculation instant. What they didn’t do was choose the assumptions. A model built on a bad growth rate produced bad answers faster, and with more decimal places.
The useful change wasn’t buying the software. It was moving people from calculating to questioning: what if sales slip a quarter, what if the rate changes, which number are we least sure of.
The business decision: automate the repeatable part, then put the time saved into the part that needs judgment. If you skip the second half, you’ve only made your mistakes cheaper.
Desktop publishing made layout cheap
In 1985, Aldus PageMaker, running on the Apple Macintosh and printing to the new Apple LaserWriter, put page layout on a desk. Work that had gone to a typesetter could be done in an afternoon by anyone with the equipment.
What followed was a wave of newsletters and flyers with six fonts on one page. The tool removed the cost of production. It didn’t supply typography, hierarchy or a sense of what the reader needed first.
Design didn’t disappear. It got more visible, because the difference between considered work and default work was suddenly sitting side by side on the same office noticeboard.
The business decision: owning the tool isn’t the same as having the skill. When everyone can produce, the advantage moves to whoever knows what good looks like.
Flipping the script only works if you understand the story first.
Where AI fits the same pattern
AI makes first drafts, variations and summaries cheap. Some of that is a real improvement. Google’s responsive search ads, for example, already take multiple headlines and descriptions and test combinations. Drafting a wider set of options to choose from is a sensible use of AI. Deciding which ones are accurate, on-brand and worth the budget is still a person’s job.
The same goes for content. AI can tidy an interview transcript into a structured draft. It can’t know what your senior estimator learned on the last ten jobs. That’s why AI content without your expertise reads like everyone else’s.
Diagnose the work before you automate it
Before you hand a task to AI, separate four things that often get bundled together.
| Question | What to ask | Example: monthly client reports |
|---|---|---|
| Speed | Does doing it faster matter? | Reports go out days late; faster helps |
| Cost | What does it cost now, including review? | Hours of senior time pulling numbers |
| Quality | Would faster or cheaper make it worse? | Summaries must match the data exactly |
| Judgment | Who decides what it means? | The account lead explains why and what next |
Say you run a 20-person engineering consultancy that sends clients a monthly project report. Pulling the figures and drafting the summary takes a senior engineer half a day each time. AI can help assemble the draft. The engineer still checks the numbers and writes the two paragraphs that matter: what changed and what the client should decide. The time saved goes to the conversation, not just the inbox.
Name the task.
One specific piece of work, not “marketing” or “content”.
Find the judgment in it.
Which part depends on experience, context or a decision? That part stays with a person.
Automate the rest, then review.
Count the review time too. A fast draft that needs an hour of fixing isn’t fast.
Check the outcome, not the output.
Did the work get better for the client or buyer, or did you just produce more of it?
If you’re working through where AI belongs in your marketing as a whole, that’s a strategy question before it’s a tooling one.
Every tool that made work cheap made judgment more valuable.
AI won’t be the exception.
Decide what stays human before you decide what to automate.