Do you know which is which?
Situational. It will change what they do. You may or may not need fewer of them.
Current AI tools are very good at drafting, summarizing, categorizing, and searching. They are not good at judgment, relationships, accountability, or anything requiring context that lives in someone’s head. What actually happens in practice: employees get faster at the boring parts and have more time for the work that actually requires a human.
AI drafts the first version of emails, proposals, and reports. People review and send.
AI summarizes long threads, documents, and call transcripts. People act on the summary.
For reference: ATMs did cut the number of tellers a branch needed. Branches got cheaper to run, so banks opened more of them, and ended up employing about as many tellers as before.
False. (You don’t understand your microwave either.)
You don’t need to know how large language models work to use ChatGPT any more than you need to understand internal combustion to drive. The tools are designed for non-technical users. The learning curve is real but it’s measured in days, not months.
Start with one task you do every week that involves writing or searching.
Use the tool for that one thing until it feels natural.
Add a second use case. Repeat.
True — which is why your employees need to stay in the loop.
“Hallucination” is a real problem. AI tools do sometimes state things confidently that are wrong. The answer is not to avoid AI — it’s to use it for tasks where a human is reviewing the output before anything happens. Use AI as a first draft, not a final authority.
High-trust tasks (legal filings, financial statements, medical advice) — do not rely on AI output without expert review.
Lower-trust tasks (draft emails, summarizing documents, brainstorming) — AI is excellent, errors are low-stakes.
The model improves when you correct it. Treat it like a capable new hire who needs feedback.
False. Things are not settling down.
Despite confident predictions, the pace of change in AI is not slowing. Waiting for stability is waiting for something that isn’t coming. The businesses building AI habits now — even small ones — will have a real operational advantage over those who wait. The good news is that starting small is genuinely the right approach, so there’s no need to wait until you have a “strategy.”
One person, one task, one week. That’s a sufficient start.
Your competitors who are using AI are mostly using it for the same mundane things — drafting, summarizing, searching.
The gap compounds over time. Start accumulating the habit now.
False. Unless you train it yourself.
AI reacts to your data. It does not absorb it.
AI is not quietly studying your business in the background. It only knows what you put in front of it, when you put it there.
If you want continuity, you have to create it — saved prompts, documents, or systems that feed it the same context repeatedly.
Assume anything you paste could leave your control. As with any third party, use good judgment.
What it’s good for
| Task | What AI does | Human still needed for |
|---|---|---|
| Email drafting | Writes a professional first draft from a few bullet points or a rough note | Tone, relationship context, final send decision |
| Meeting summaries | Transcribes and summarizes calls, extracts action items | Validating accuracy, assigning ownership |
| Document Q&A | Answers questions about contracts, manuals, or policies by reading the document | Legal interpretation, consequential decisions |
| Data extraction and manipulation | Pulls fields out of invoices, statements, and PDFs, reshapes messy exports, and flags likely duplicate records | Verifying totals and dates, and approving merges before anything is written back |
| Writing and reviewing code | Drafts working code, explains code someone else wrote, and catches obvious errors in a change | Architecture, security and permissions, testing before production |
| Troubleshooting | Reads error messages, logs, and settings to suggest likely causes and what to check next | Confirming the cause, and judging what a fix will affect before making it |
| Customer support drafts | Suggests reply text based on ticket content and past responses | Empathy, escalation judgment, complex cases |
| Research and summarization | Quickly synthesizes information from multiple sources on a topic | Fact-checking, source verification, decisions |
| Task | Why AI falls short today |
|---|---|
| Autonomous decision-making | AI can recommend, but consequential decisions require human accountability and context it doesn’t have. |
| Replacing sales relationships | Buyers still buy from people they trust. AI can support the process; it can’t substitute for the relationship. |
| Legal, tax, or medical advice | Hallucination risk is too high in high-stakes domains. Use AI to prepare questions, not to get answers. |
| Fully automated customer service | Works for simple, predictable queries. Fails visibly on anything complex, emotional, or novel. |
| Replacing your institutional knowledge | AI doesn’t know your clients, your history, or why things are done the way they are. That lives in people. |
Readiness signals
Not a strategy, not a budget. These are the conditions under which a first AI project tends to work.
There is a specific task that eats hours every week.
Someone can say what a better outcome would look like.
The work involves writing, searching, or summarizing.
A person will review the output before it goes anywhere.
The records that task depends on are in one place and mostly correct.
If most of those are true
Ten lessons from real implementations, and the five layers that have to hold before the tool matters.
Read the 10 lessons →Want to know what AI could do for your business specifically?
We practice what this page preaches: our AI use, access, and oversight policy is public, and our AI-assisted deduplication case study shows where AI actually earns its keep.
Schedule a Free Consultation