AI Agents Are Getting Smarter. Are Your Employees Getting Lazier?
AI does not make employees lazy by itself, but it can make them stop checking. The research so far points the same way: generative AI lifts output, especially for less experienced staff, while people who trust it too much think less critically about what it produces. The difference between augmentation and deskilling is not the tool, it is the policy around it.
This post weighs the evidence on both sides, gives SourceIT's verdict, and sets out a simple policy any Singapore SME can adopt this quarter. Hardware gets a light touch at the end, including AI-ready business laptops such as the ASUS ExpertBook P3 and HP EliteBook X G1i, the ASUS Ascent QN10 mini PC and the ASUS Ascent GX10 for local AI.
What does the research actually say about AI and thinking at work?
Three studies come up in nearly every version of this debate. They measure different things, so it helps to read them side by side.
| Study | Who was studied | What it found | What it does not prove |
|---|---|---|---|
| Lee et al., "The Impact of Generative AI on Critical Thinking" (Microsoft Research, CHI 2025) | 319 knowledge workers sharing 936 real examples of GenAI use | Higher confidence in GenAI was associated with less critical thinking; higher self-confidence with more. Critical thinking shifted towards verifying, integrating and stewarding AI output | It is self-reported survey data, and it comes from a company that sells AI assistants |
| Brynjolfsson, Li and Raymond, "Generative AI at Work" (NBER Working Paper 31161) | 5,179 customer support agents | Issues resolved per hour rose 14% on average and 34% for novice and low-skilled workers, with minimal impact on experienced staff | It measures output in one job type, not long-term skill |
| Kosmyna et al., "Your Brain on ChatGPT" (arXiv preprint) | 54 participants writing essays, with brain activity measured | The LLM-assisted group showed the weakest brain connectivity and the lowest sense of ownership of their essays | It is a small, not yet peer-reviewed lab study of essay writing, not office work |
Put together, the evidence supports neither panic nor complacency. AI clearly helps people produce more, and the biggest gains go to the least experienced. But the same tools can quietly remove the effort through which people learn, and over-trust is the warning sign.
The case that AI is making people lazier
The worry is reasonable. When a draft email, a spreadsheet formula or a meeting summary appears in seconds, the temptation is to accept it and move on. The Microsoft Research paper describes this directly: the more people trusted the AI, the less they reported thinking critically about the result. In the "Your Brain on ChatGPT" preprint, participants who leaned on an LLM felt the least ownership of the essay they handed in.
For a business, that can look like a proposal sent with a confident but unchecked figure, or a new hire who never learns the pricing model because the assistant always fills it in. These are small erosions of judgement that only show when the AI is wrong, unavailable or out of its depth.
Agents raise the stakes. A chatbot drafts; an agent acts. When an AI agent can read an inbox, update a catalogue or send a quotation, the human step that used to force a second look can disappear entirely. We cover the risk side of that in Your AI Assistant Might Be Your Biggest Security Risk.
The case that AI is making people better
The other side has strong evidence too. The NBER study of customer support agents is one of the largest real-world measurements available, and the headline is that the least experienced workers improved the most. The authors suggest the tool spreads the practices of top performers to newer staff, helping them gain experience faster. That is augmentation, not laziness.
Spreadsheets and spell-check triggered the same argument once. What matters is whether staff still understand what the tool is doing and can spot when it is wrong. The Microsoft Research finding can be read optimistically as well: critical thinking did not vanish, it moved towards verifying information, integrating responses and taking stewardship of the task. Those are skills a business can train.
AI does not make people lazy. Unreviewed AI does. Keep a named human responsible for every output and the tool becomes a coach instead of a crutch.
SourceIT's verdict: augmentation, if you design for it
We think the "lazy employee" framing misses the point. Most staff use AI because they are busy, not because they are idle. The real risk is unowned work: output that nobody has checked and nobody feels responsible for. When that happens, mistakes leak out and juniors stop building the judgement that the business will need from them in five years.
So our position is simple. Use AI widely for first drafts, summaries, research and repetitive admin. Keep humans in charge of facts, numbers, commitments to customers and anything with legal or financial weight. And make sure junior staff still do some work the hard way, so they can tell when the AI has gone wrong. If you are still at the pilot stage, our post on why AI projects look amazing in demos but fail in real offices covers the traps.
A practical AI policy that keeps people sharp
A one-page policy beats a fifty-page one that nobody reads. These five rules cover most SMEs:
- Ownership: every AI-assisted output has a named owner who signs it off. "The AI wrote it" is never an answer to a client.
- Review by risk: internal notes can go out lightly checked; prices, contracts, technical specifications and anything sent to customers or regulators get a full human review against the source.
- Show your sources: staff should be able to point to where a figure came from. If the AI cannot cite it, a person verifies it.
- Learn before you delegate: new joiners do core tasks manually first (a quotation, a reconciliation, a report) before handing them to an assistant.
- Approved tools only: decide which AI services may see company data and which may not. Our guide to where your company data goes when you use AI explains what to check, and Are Businesses Becoming Too Dependent on ChatGPT? covers the single-vendor risk.
Training matters as much as rules: short, regular sessions on fact-checking beat a long annual course. Ask staff to share one AI output they caught being wrong each month; it normalises checking instead of blind trust.
Does the hardware make a difference?
Only a little, and it should not lead the conversation. A modern business laptop keeps everyday AI features and cloud assistants responsive, and newer "AI PC" models add on-device acceleration for some Windows and vendor features. Examples on our store, as listed on 23 Oct 2026:
- ASUS ExpertBook P3 P3406CCA-LY0040X (Intel Core Ultra 7 255H, 16GB, 512GB SSD): S$2,099, in stock.
- HP EliteBook X G1i 14 U5-236V 16GB 512GB AI Notebook: S$2,584, available to order.
- ASUS Ascent QN10 AI Mini PC (Snapdragon X2 Elite, 32GB, 1TB, Windows 11 Pro): S$2,375, in stock.
Where hardware does change behaviour is local AI. Running models on an in-office box such as the ASUS Ascent GX10 (128GB / 1TB), S$7,200 and in stock, lets you decide which data the AI may touch and keep it inside your network. That makes it easier to roll AI out to everyone under one set of rules. Read when running AI locally beats the cloud and the AI employee on your desk for the trade-offs, or browse AI workstations and AI servers. If you are weighing how far to push agents, see One Employee, Ten AI Agents.
Buying AI-ready PCs and local AI systems in Singapore
SourceIT is an ASUS reseller and also supplies HP with local warranty; the ExpertBook P3 listing states 3 years of local onsite warranty, and for other items confirm the term for your SKU when requesting a quote. Prices are as listed on 23 Oct 2026. For team roll-outs ask about volume discounts and special pricing. Use Request for Quotation on the product page, call (65) 6978 3502 or email Sales@sourceit.com.sg. Government buyers can purchase through GeBIZ.
Frequently asked questions
Does using AI make employees lazy? Not by itself. Research from Microsoft Research found that people with higher confidence in GenAI reported less critical thinking, so the risk is over-trust. Clear ownership and review rules keep staff engaged.
Does AI improve productivity at work? In the NBER study of 5,179 customer support agents, AI assistance raised issues resolved per hour by 14% on average and 34% for novice and low-skilled workers, with minimal impact on experienced staff.
What is AI deskilling? It is the loss of skills people stop practising because a tool does the work for them. It matters most for junior staff who have not yet built the judgement to spot AI mistakes.
How do I stop staff over-relying on AI? Name an owner for every output, require review in proportion to risk, ask for sources behind figures, and have new joiners learn core tasks manually before delegating them to AI.
Is the "Your Brain on ChatGPT" study conclusive? No. It is a small arXiv preprint with 54 participants writing essays. It is a useful signal about engagement, not proof of long-term harm in office work.
Do we need AI PCs to use AI at work? No. Most AI assistants run in the cloud on any modern laptop. AI PCs help with some on-device features, and a local AI system such as the ASUS Ascent GX10 helps when you want to keep data in-house.
