Part 1: Brains on: how to work with AI without switching your thinking off 

AI at WorkLeadershipWays of Working
Written By Catherine Russell

A quick look at the landscape

We’re moving fast. New tools land every week. It feels like when we all jumped on Facebook before we really understood that “public” meant… well… public. We shared first and read the settings later. With AI, we risk the same thing: rushing in before we understand rules and our habits catch up. Back then, defaults nudged us to “share to everyone.” Today, defaults nudge us to “accept the suggestion.” Different tech, same pattern.  

AI is already part of work. Recent surveys suggest roughly three in four knowledge workers use AI. Many bring their own tools and keep quiet because they worry it makes them look replaceable. Leaders want AI skills, but many still lack a plan; they are still figuring AI out themselves, let alone how to get their team skilled. Power users say AI helps them save time, focus and feel more creative. Without outsourcing all the fun thinking parts, that's the upside we want. Also, BYOAI is risky: unapproved tools can leak company data and customer information.   

The rulebook is still being written.

Europe has passed the first broad, cross-sector AI law (the EU AI Act). A quick overview - it bans a few uses (like social scoring), puts stricter duties on high-risk AI (safety, compliance), and asks for basic transparency (e.g. label deepfakes, be clear when people are chatting with an AI). It’s a phased rollout through 2025–27, so it won’t answer every day-to-day question yet.

In Australia, the government has issued an interim response and practical toolkits while the full details are worked out (links to these resources below).

Meanwhile, lawsuits and regulatory probes are testing the boundaries of copyright, privacy, safety, and more, so AI providers keep updating their terms and product rules as they learn.

The bottom line: don’t wait for perfect rules. Set sensible boundaries now and keep an eye on vendor terms, they change!

Note: This post is general information, not legal advice. Check your organisation’s policies and local laws, and review vendor terms with your legal/privacy team.

AI autopilot in action

AI landed fast. Everyone’s trying to use it, and we should. But while we figure out the norms and some of us fumble through adoption, we still need to flex our thinking alongside the tools.

A quick snapshot of “speed without intent”
A support team has had 50,000 tickets since last month, which is well above their usual numbers. They paste, “Cluster our tickets and tell us the top problem,” into an AI tool and implement fixes based on the first answer. The team rolls out a complex new process and a long set of tooltips. Problem solved. (NB Data safety: Strip personal/confidential data first or run analyses in your enterprise workspace (no data retention, private endpoint).

What went wrong?
They optimised for speed to an answer, not speed to a good answer.

The goal and audience weren’t set. Source/date checks and a small sample review were skipped, so the AI clustered a short-term spike rather than the main driver. The team shipped a complex process that didn’t fix the cause. Repeat tickets didn’t drop, and customer confidence slipped.

So what fixes “speed without intent”? Definitely not a bigger policy or more meetings, just five tiny moves that keep you in charge of the thinking while AI does the heavy lifting.

Five Checks to Keep Thinking On (and stop “the tool said so”)

These checks take minutes and add just enough structure.

Data note: Just to be crystal clear, don’t paste customer or employee PII and commercially sensitive data into public models. Redact it (names, emails, phone numbers, addresses, IDs, free-text notes) or use your company’s approved enterprise model (no data retention, private endpoint, correct region).

  1. Frame the decision before you prompt
    In one line, say what you’ll change, who it helps, and how you’ll know it worked.
    Example: Add a short ‘next steps’ note to replies for first-time customers; success = repeat emails down 20% in 14 days.”

  2. Cross-check, don’t copy-paste
    Run two prompt variations and hand-skim a small sample of real entries. Check one internal signal (e.g., agent tags, logs, CSAT) and one external signal (e.g., forum threads, reviews, news), ask for a confidence score. If there’s no visible evidence, don’t use it.

  3. Check the source and date on anything that impacts risk, money, or customers
    Write down the source and the date range you analysed, note what’s missing, and call out any unusual events that could skew results (e.g., promo, outage, big release). If you can’t see the source/date, don’t use it.

  4. Run one counter-prompt
    Try: “What else could explain this spike?” “Five ways this fix could backfire.” “What’s the simplest change with the biggest upside?” You’ll catch the issues before customers do.

  5. Make a human decision and log the why
    Write one line with: Decision | Owner | Why it should work | Success signal | Review date

Result: AI handles the volume while teams keep the thinking.

And the support team? They got smart and used the guardrails. In a two-week pilot, the team tested small changes, simplified the flow, and made the next step clear. Time-to-insight dropped from weeks to days, repeat tickets fell, CSAT lifted, and the team won hours back. It was better for customers and the team. Win Win!

Why this keeps your brain on

Clear goal → focus. A one-line goal and success signal stops “apoints attention at the real job and keeps AI on task.

Cross-check → confidence Two variations, one outside and one inside signal, plus a small human sample, give you a second opinion and catch pattern mismatches.

Source & date → evidence. Note where the data came from, the date range, and any unusual events, so short spikes don’t drive big decisions.

Counter-prompt → healthy scepticism. Ask “What else could explain this?” and “How could this go wrong?” to surface blind spots early.

Human decision & why → faster learning next time. Log the decision, owner, why, success signal, and review date so you learn faster and stay accountable.

Together these habits keep your teams thinking engaged and reduce cognitive load, while AI handles more of the heavy lifting at scale.

Next up — Part 2: The Thinking modes - Fast, practical drills to keep your best thinking switched on—AI as a power-assist, not a replacement.

Resources & further reading

AI at work (usage & skills)

Education guidance (use it with rules)

What the law is doing (and when)

Behavioural risks to watch (why “brains on” matters)

Productivity effects & early brain findings

Catherine Russell
Previous

Part 2 - The Thinking Modes: keep your brain strong in the age of AI
Next

Before we go faster: protecting our thinking in the age of AI