Protect the Practice
Which AI use removes the practice that builds judgement, and which use protects it
This essay sits primarily under the Can we do AI? question of the practice’s methodology: the cognitive consequence of how a firm uses AI, and which uses protect the practice that builds judgement.
The question behind the question
Every principal who has watched a graduate paste a question into a model and accept the first answer has felt the worry. Is this making my team dumber?
The honest answer is that the question is badly aimed. AI does not act on a person’s mind in a fixed way. The same tool can remove the practice that builds judgement or protect it. The variable is the use, and a firm can design the use.
So the useful question is narrower. Which uses remove the practice that builds memory, judgement and skill, and which uses protect it?
That question has an answer a principal can act on. It also has a claim the evidence can support.
What cognitive offload actually is
Cognitive offload is using an external tool to reduce mental effort. A shopping list, a calculator, a map, a reminder. None of these make a person dumber. They free attention for something else.
AI is the most powerful offload tool we have used, because it can take the drafting, analysing and evaluating as well as the recall. In a regulated practice, the thinking is the work, so the stakes are higher than they were with a calculator.
Passive delegation and active use
Passive delegation accepts the output without framing the problem, checking the claim, or making the decision. The work is completed. The practice is not.
Active use puts the effort where it builds capability. A rough pass before the comparison. Alternatives that challenge the first answer. Shared work that widens the table. Better questions before the conversation. A person who keeps the decision and signs it.
The tool is the same in both cases. The use is what differs.
Conversation, Not Delegation draws the line this way. An output request can be appropriate. If the purpose is to learn, preserve occasions to attempt, explain and test. The delegation trap is accepting a result you cannot defend.
The mechanism
The mechanism is not struggle for its own sake. It is the right amount of challenge, with feedback and a way to test whether the capability transfers.
If AI removes every difficult step, the reps that build memory, judgement and skill disappear. If it takes over the low-value production and leaves time for framing, checking, deciding and connecting, it can protect the capacities people worry about. The saved time has to go somewhere useful, or the efficiency trap fills it.
The Human Edge makes the same point about the team. When AI takes the easy work, juniors lose reps they never chose to skip, and somebody has to put the practice back on purpose.
What the evidence says
The evidence is more specific than the headline.
A 2025 survey of knowledge workers by Lee and colleagues found associations between generative AI use and self-reported critical thinking and checking practices. It does not measure lost intelligence.
Bastani and colleagues studied AI assistance in high school mathematics and found that unguarded use can harm learning. The design of the help mattered. Their result does not settle what happens in a law firm or an accounting practice.
Neither study licenses the claim that one copied answer permanently damages a person’s mind. Read each one at the level it supports.
What practice disappears
Four losses are worth watching in a professional team.
1. Productive struggle
Some difficulty is professionally valuable. When a graduate wrestles with a problem and breaks through, the struggle builds durable judgement. If AI eliminates all of it, it may eliminate the capability that comes with it.
The response is to structure work where struggle is built in. Require an attempt before AI consultation, then reflect on what AI added.
2. Illusion of understanding
Staff may feel they understand something because they have read a clear explanation. They can paraphrase it but cannot apply it to a new client or a new set of facts.
The response is to test transfer rather than recall. Ask staff to apply the concept to a novel situation and explain the reasoning aloud.
3. Erosion of foundational skills
Some skills underpin higher-order thinking. If staff skip the foundations, they may struggle with the complex tasks that depend on them.
The response is to be intentional about which foundations matter. Not every traditional skill is equally foundational. Protect the ones that genuinely underpin the work and let go of the ones technology has superseded.
4. Metacognitive atrophy
Staff stop monitoring their own understanding because the output gives them a false sense of competence. They stop asking “do I understand this?” and start asking “does this look right?”
The response is to build reflection into the work. Ask what they understood before AI, what they understand now, and where AI helped or misled them. Make the thinking visible.
What protects the practice
The same principle applies at every level. AI does not inherently support or weaken capability. Supervision and work design do.
For a team, the practices that protect capability look like this:
- Staff record how AI was used, what they changed, and why.
- Staff assess AI output against criteria, identify errors, and improve it.
- Before signing or sending, staff explain the reasoning in their own words.
- Hard cases get attempted without AI first, then with AI, then compared.
- Reflection is built in: what did I understand before, and what do I understand now?
- AI is used as a thinking partner rather than an answer machine.
The common thread is that the person stays in the loop. Approval after the fact is not enough. They think alongside the tool while the work is being done.
The claim to make
It is tempting to turn this into a slogan. Do not.
Say that a well-designed use of AI can protect against the loss of practice that comes from passive delegation. Do not say that AI prevents brain rot. The first claim is defensible and specific. The second is a marketing line that a sceptical partner will test in the first meeting and find empty.
The use matters more than the tool. That sentence will survive scrutiny. Most of the louder ones will not.
What to do
For supervision:
- Design for process, not just product. Make the thinking visible.
- Build in reflection points: what did you understand before AI, and after?
- Test for transfer and explanation, not just a correct output.
- Protect genuinely foundational skills and release the ones that no longer matter.
- Model your own use. Show the team how you think with the tool.
For your own practice:
- Notice when you are offloading and when you are collaborating.
- Do hard thinking without AI often enough to keep your own foundations.
- Use AI to extend your capability, not to replace your engagement.
- Stay curious about what your team is building, not just what it is producing.
For the conversation with peers:
- Acknowledge the legitimate concern. Cognitive offload is real.
- Move from “AI makes us dumber” to “how do we design for thoughtful use?”
- Share what works.
- Resist uncritical enthusiasm and reflexive fear in equal measure.
The bottom line
The question is not whether AI makes us dumber. It is which use removes the practice that builds memory, judgement and skill, and which use protects it.
Cognitive offload is not the enemy. Cognitive abdication is. The difference is whether agency stays with the practitioner.
Protect the practice. The person who keeps the decision is the person who keeps the capability.
About this work
borck.consulting helps regulated Australian small firms use AI without hollowing out the judgement that makes the work defensible. Engagements begin with the AI Readiness Diagnostic.
The argument here is set out at book length in Conversation, Not Delegation and The Human Edge. The passive use distinction comes from Conversation, Not Delegation. The team version, where the reps disappear, comes from The Human Edge.
Further reading
- Lee, H., Sarkar, A., Tankelevitch, L., Drosos, I., Rintel, S., Banks, R., & Wilson, N. (2025). The impact of generative AI on critical thinking: Self-reported evidence from a survey of knowledge workers. https://www.microsoft.com/en-us/research/wp-content/uploads/2025/01/lee_2025_ai_critical_thinking_survey.pdf
- Bastani, H., Bastani, O., Sungu, A., Ge, H., Kabakcı, Ö., & Mariman, R. (2024). Generative AI without guardrails can harm learning: Evidence from high school mathematics. Proceedings of the National Academy of Sciences, 121(26), e2422633121. https://doi.org/10.1073/pnas.2422633122
- Risko, E. F., & Gilbert, S. J. (2016). Cognitive offloading. Trends in Cognitive Sciences, 20(9), 676–688. https://doi.org/10.1016/j.tics.2016.07.002
- Sparrow, B., Liu, J., & Wegner, D. M. (2011). Google effects on memory: Cognitive consequences of having information at our fingertips. Science, 333(6043), 776–778. https://doi.org/10.1126/science.1207745
- Gerlich, M. (2025). AI tools in society: Impacts on cognitive offloading and the future of critical thinking. Societies, 15(1), Article 6. https://doi.org/10.3390/soc15010006
- Fan, Y., Tang, L., Le, H., Shen, K., Tan, S., Zhao, Y., Shen, Y., Li, X., & Gasevic, D. (2025). Beware of metacognitive laziness: Effects of generative artificial intelligence on learning motivation, processes, and performance. British Journal of Educational Technology, 56(2), 489–530. https://doi.org/10.1111/bjet.13544
- Dellermann, D., Ebel, P., Söllner, M., & Leimeister, J. M. (2019). Hybrid intelligence. Business & Information Systems Engineering, 61(5), 637–643. https://doi.org/10.1007/s12599-019-00595-2