Old IT Was Built for Humans. AI Needs Something Else.
Old IT started with the mainframe.
IBM didn’t replace people. It changed what work meant.
Things that took rooms full of clerks could now run overnight on one machine.
Then came PCs. Microsoft and Intel gave us Word, Excel, and PowerPoint.
For many jobs, this felt like magic. Change one cell in Excel, and everything recalculates. That was powerful.
Then came the cloud. SaaS.
We started collaborating in real time.
Figma. Canva. Google Docs. HubSpot.
Agile replaced waterfall. Central IT teams formed with a clear goal: don’t break things and keep costs under control.
This was Old IT.
It had a clear shape.
It was layered:
CPU → OS → desktop apps → browser → web apps.
It was siloed:
Sales lived in Salesforce.
Finance in SAP.
HR in Workday.
Tickets in ServiceNow.
Each system owned its territory.
Most importantly, it was built for humans.
Think about a typical finance workflow.
Someone builds a model in Excel.
Creates slides in PowerPoint.
Exports to PDF.
Emails it for review.
XLS. PPT. DOCX. PDF.
These formats exist because humans need to see work.
PDF is great for human eyes.
It’s terrible for machines.
We are deeply anchored in this world:
files, folders, screens, permissions, vendor boundaries.
Then the shift happened
ChatGPT showed up, and suddenly everything felt different.
But here’s the problem.
We’re still using Old IT.
Every vendor is adding a chat box.
Then an “AI assistant.”
Then an “agent.”
But it’s still the same software underneath.
Old IT shows you a screen and waits for you to act.
It displays work. It doesn’t do work.
Something deeper needs to change.
A glimpse of something new
Anthropic’s Claude Code gave a small preview.
No dashboards.
No slides.
No spreadsheets.
Just instructions in. Work out.
It works well for code because code already lives in formats machines understand: text, packages, docs.
But outside of code, we hit a wall.
PDFs.
Spreadsheets.
Access controls.
Silos.
So we build workarounds: document parsing, MCP servers, plugins.
Helpful, yes.
But still workarounds.
Why Old IT exists (and why it won’t vanish overnight)
Before blaming Old IT, we need to be honest.
PDFs exist for a reason.
Contracts need to look the same everywhere.
Reports need to be auditable.
Access controls exist for a reason.
Some data cannot leak. Ever.
Vendor silos exist because integration is hard.
And standards are slow.
Old IT is not stupid.
It is the result of decades of real problems being solved.
The question is not “delete it all.”
The question is: what can now be done differently?
What New IT looks like
If we strip things down to first principles, what do we really need?
We need work to get done.
Not slides.
Not files.
Not formats.
Work.
The key shift is this:
- Humans stop operating software
- AI operates software, humans direct it
You stop clicking.
You start instructing.
Other changes follow naturally:
- Formats optimized for display → formats optimized for action
- Files as endpoints → structured data until the last moment
- Apps handing work off → AI holding context across systems
- Permissions on files → permissions on actions
- Workflows in UI → workflows in language
What this could mean in practice
No more pitch decks as the primary artifact.
You describe the deal. The AI pulls data and creates what each audience needs.
No more spreadsheets as the calculation engine.
The agent runs numbers directly on the data source.
The spreadsheet becomes a view, not the workspace.
No more PDFs as the system-to-system format.
Data stays structured.
Human-readable output is generated only when needed.
Access control shifts from “can you see this file?”
to “can the AI do this action for you?”
The transition will be messy
This won’t happen overnight.
Old IT has years of infrastructure, regulation, and habit behind it.
The shift will be gradual.
But the direction is clear.
AI doesn’t fit well into Old IT.
The formats are wrong.
The interaction model is wrong.
We need a New IT.
Slowly, quietly, it’s already coming.
Who will win
Right now, there are two ways to build AI products.
One way:
Add AI to existing software.
Chat on top of CRM.
Agents bolted onto old workflows.
This captures the transition.
The other way:
Build AI-native from the start.
Design for instructions, not clicks.
For actions, not screens.
This captures what comes after.
Adding AI to Old IT is like adding voice control to a typewriter.
Helpful — but still the same machine.
AI-native is the word processor moment.
Once people feel that shift, they won’t go back.
Final thought
This isn’t the death of IT.
And it’s not the death of SaaS.
It’s the end of software that exists mainly to be clicked.
The future belongs to systems that execute, not display.
Old IT got us here.
New IT will take us forward.