AI at work is starting to look like a contact sport. On one side, people (often non-technical) who run agents, wire tools together, automate workflows, and save hours every week. On the other, entire teams who “chat” with ChatGPT like it’s a polite search engine.
That’s not a small difference. It’s a structural productivity gap.
Martin Alderson put it plainly: two kinds of AI users are emerging, and the gap is astonishing. We see the same thing at Deepthix every day. The difference isn’t IQ, job title, or even technical skill. It’s how you use AI.
- define the two profiles (no fluff),
- explain why classic enterprise setups are sabotaging themselves,
- give you a practical 30-day plan to move from “chat” to “automation”.
The two camps: power users vs chat-only users
1) Power users: AI as an operator, not a toy
Power users don’t “ask questions.” They assign missions.
- CLI agents (Claude Code, Copilot CLI, etc.),
- workflows (Make, n8n, Zapier, Python scripts),
- connectors (MCP, APIs, webhooks),
- environments where AI can act: read/write files, run tests, transform data, call services.
Counter-intuitive detail: many aren’t very technical. Alderson notes he’s seen lots of non-devs using Claude Code in the terminal for non-software tasks. Same in the field: finance, ops, HR, sales ops… anywhere you have spreadsheets, rules, exceptions—AI + a real execution environment becomes a weapon.
- 200 PDF invoices/month,
- extract amounts/VAT/due dates,
- reconcile with bank lines,
- push into your accounting tool.
A chat-only user asks for a “template.” A power user builds a flow: extraction → validation → export → alerts. Result: hours saved, fewer errors, and an auditable process.
2) Chat-only users: AI as a nicer Google
- they use ChatGPT (or similar) conversationally,
- they copy-paste text,
- they prompt “write me an email / summarize this / brainstorm ideas.”
It helps. But it caps fast.
- no permission to install tools,
- no access to internal data,
- no clear experimentation process,
- and corporate tools that are… mediocre.
The real accelerator: execution (not conversation)
The core difference: the ability to execute.
- Chat-only users stay in text land.
- Power users connect text to reality: files, databases, CRM, billing, support tickets, code.
That’s why “directive AI” (precise instructions + execution) is rising in professional settings. Reporting on Anthropic’s work (via Windows Central) highlights that a large share of API-driven tasks are heavily automated—more operational than conversational.
Translation: profitable AI is AI connected to your systems.
Microsoft 365 Copilot: when the bundle slows you down
Alderson is blunt about M365 Copilot: weak interface, laughable “agent” feature, limited execution, slow performance, and poor handling of larger files.
This isn’t just complaining. It’s a pattern.
- imposed by IT,
- bought as a bundle,
- evaluated via slides,
…it becomes politically viable minimum, not best-in-class.
The most telling point: Alderson notes Microsoft is rolling out Claude Code internally, despite having Copilot at near-zero marginal cost and deep ties to OpenAI. If even the vendor doesn’t rely solely on its own tool, you can infer the gap.
Why this is dangerous for enterprises
Because leaders try AI through a constrained tool → get mediocre results → conclude “AI is overrated.”
Meanwhile, more agile competitors automate for real.
- In the US, enterprise adoption of paid AI services reached 46.6% (Ramp data reported by Business Insider, Dec 2025), with OpenAI leading (36.8%) ahead of Anthropic (16.7%).
- But daily usage is still chilly: a late-2025 Gallup survey (reported by TechRadar) suggests 49% of US workers say they never use AI in their role, and quarterly growth is small.
Bottom line: AI is in budgets, not yet in habits—and that’s exactly where the gap widens.
The real culprit: “zero-risk” IT policy (therefore zero progress)
- locked-down environments (no Python, no scripts, sometimes even VBA is restricted),
- siloed data and absurd permissions,
- procurement that takes months,
- “only one approved tool” (often the weakest one).
Result: showroom AI, not production AI.
Then come overpriced consultancies selling “AI transformation programs” that deliver… POCs that never ship. Classic corporate theater.
How to move from chat-only to power user in 30 days
You don’t need to be a developer. You need a plan.
Week 1 — Pick one bleeding process (and measure it)
- inbox triage and replies,
- lead qualification,
- weekly reporting,
- PDF/Excel data extraction,
- CRM updates.
Measure: hours/week, error rate, cycle time.
Week 2 — Give AI an execution environment
Goal: stop copy-pasting.
- n8n (self-hostable) to orchestrate,
- Make/Zapier to move fast,
- a small Python script (even on a server) for file handling,
- API connectors to your CRM/ERP.
Key point: AI must be able to read inputs and write outputs without you.
Week 3 — Add guardrails (or you’ll get burned)
Automation ≠ chaos.
- human approval for risky actions (payments, deletions, external sends),
- logging (who did what, when),
- sample-based testing,
- exception handling (when uncertain → escalate).
Week 4 — Industrialize and deploy
Move from “works on my laptop” to “runs every week.”
- one-page documentation,
- monitoring (basic: Slack/email on failure),
- clean access rights,
- before/after KPIs.
- hours saved,
- cycle time reduced,
- errors avoided,
- cash freed.
Use cases that create a real gap
Ops / Admin - auto-generate contracts from CRM fields - payment reminders personalized by status - create internal tickets from incoming emails
Sales - prospect research + scoring (site, LinkedIn, signals) - outreach emails tailored to industry + objections - auto-update CRM after calls (summary + next steps)
Finance - multi-source consolidation (bank, invoices, expenses) - anomaly detection (duplicates, unusual amounts) - monthly reporting automated (charts + narrative)
Not sexy. Profitable.
The near future: a durable bifurcation
Consumer numbers are massive: Menlo Ventures estimates 1.7–1.8B people used an AI tool in the last 6 months, but only 3–5% pay. That implies huge reach, often shallow usage.
In parallel, enterprise spend concentrates on integrated usage (APIs, agents, automation). That’s where durable advantage is built.
So yes: two kinds of AI users are emerging.
The question isn’t “do you use AI?”
It’s: do you let it talk, or do you make it work?
Want to automate your operations with AI? Book a 15-min call to discuss.