I do not believe AI's main impact will be replacing whole teams. In the work I see, its immediate value is removing tasks that keep experienced people busy.
Some jobs and tasks will change or disappear. I do not see one answer that applies to every role, company, and industry. The Stanford AI Index 2026 includes positive, neutral, and negative productivity results depending on the task and worker context.
So I would ask a narrower question: which repeated work can disappear, and what can your experienced people do with the time?
What I see in my own work
One change matters to me: a product manager, or anyone close to the problem, can turn an idea into a rough visual prototype before working through it with design and other experts. The idea and decision remain theirs. AI shortens the path from a problem statement to something the team can see and discuss.
I used AI in a team retrospective for the same reason. It clustered and visualised the feedback during the session. Five minutes in, we had a view of the patterns and could spend the remaining time discussing what to change.
Without that step, someone would normally collect the notes, group them, prepare the analysis, and bring it back later. AI did not decide what the team should improve. It shortened the path to the useful conversation.
Product documentation works the same way. A model can draft a PRD, epics, and user stories in minutes. The generated structure is the easy part. Someone who understands the product still has to catch the missing constraint, weak assumption, or user need.
AI handles parts of the work that consume time. People still decide whether the result is useful and correct.
Give the quiet expert time
Most organizations have someone who bridges business and technology. They understand the customer problem, know which systems are fragile, and can translate an executive request into something a delivery team can use.
They are often the same person pulled into every meeting, status report, escalation, and “quick question.” Their value is judgment, but much of their day goes into moving context between people.
Giving that person another AI chat subscription changes little. Connecting AI to one real workflow can change a lot.
If meeting notes, issue grouping, first-draft documentation, or repeated analysis becomes faster, the expert can spend more time on the decision. That is the gain I would measure.
Would they use the saved time well? That depends on the priorities and authority around them. Removing busywork without changing what the person is allowed to decide only creates a cleaner queue.
Scale stops being the only advantage
Take a small legal firm as a hypothetical example. A larger firm can put more people on document review and precedent research. The smaller firm may have excellent lawyers but less capacity for the supporting work.
If AI reduces part of that research and drafting load, the smaller firm can cover more ground. It still needs qualified lawyers to judge relevance, accuracy, confidentiality, and strategy. The tool changes the capacity gap. It does not turn legal work into a button.
I see the same opportunity wherever experts lose time preparing, copying, and sorting information. The point is to get more value from the expertise already there. It does not mean the profession has been transformed.
Do not remove the path that creates experts
A senior can use AI to work across a wider area because years of practice provide a reference for what looks wrong. A junior does not yet have that reference.
That is an argument for better mentoring and safer review, not for removing junior roles. If a company stops developing people because AI can produce the first draft, it also weakens the future pool of people able to check those drafts.
I cover the team and junior-development tradeoff in Your next team is three people and a pile of agents.
AI needs structure around it
A polished output can still be wrong. A useful workflow needs:
- a clear task and expected result
- access to the right data and tools
- boundaries for sensitive data and risky actions
- a person responsible for review
- a way to measure whether the result improved
If the source data is poor and the process is undocumented, AI can produce a confident answer from weak inputs. Faster nonsense is still nonsense.
This is why I would not start an adoption programme with a tool licence or a headcount target. Start with the work.
What I would do as a leader
- Ask experienced people which repeated task consumes time without using much judgment.
- Pick one workflow with a result you can verify.
- Remove the compliance, access, or data blocker that prevents a safe test.
- Let the person who owns the work design and review the test.
- Keep it only if the result is better after review, not merely faster to produce.
Do not begin with “where can we remove people?” That question encourages teams to hide risk and protect work. Ask where expertise is being wasted on preparation, copying, reformatting, or repeated analysis.
The harder part is giving people access to the right context while meeting IT security and compliance requirements. In my experience, those limits prevented us from using much of AI's capability or changing the workflow around it.
Start with one workflow
Pick one recurring task this week. Test where AI saves time, check the output yourself, and keep the workflow only if the result is better.
One real workflow will teach you more than another prediction about how many jobs AI will replace.
Sources
- AI Index Report 2026: Economy, Stanford Institute for Human-Centered Artificial Intelligence. Shows that measured AI productivity effects vary by task, worker, and study setting.
- Generative AI at Work, The Quarterly Journal of Economics. Reports an average productivity gain in a field study of 5,172 customer-support agents, with larger gains for less experienced workers.
- Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity, METR. Reports a slowdown for experienced open-source developers using early-2025 AI tools in a specific real-world setting.
Get Personalized Help
Copy this prompt to ChatGPT, Claude, or your favorite AI assistant. Fill in your details and get guidance tailored to your specific situation.
I'm using https://ivanmisic.net/blog/ways-of-working/ai-isnt-coming-for-your-team to remove one piece of busywork without hiding judgment or cutting the path that develops expertise. My context: - Role and recurring workflow: [WHO DOES IT, HOW OFTEN, AND THE RESULT IT PRODUCES] - Steps and time: [COLLECTING, COPYING, SORTING, DRAFTING, REVIEWING, AND DECIDING] - Data and action boundaries: [SENSITIVE INPUTS, ALLOWED TOOLS, RISKY ACTIONS, AND POLICY] - Quality and development needs: [WHO REVIEWS, HOW GOOD IS MEASURED, AND WHAT JUNIORS MUST PRACTISE] Ask for missing facts, then identify the single step that consumes time while using the least judgment. If none fits, say so. Design one small trial with: 1. a clear input, expected result, and accountable reviewer; 2. the exact boundary between preparation the tool may handle and judgment a person retains; 3. safe access to the minimum data and no authority for consequential actions; 4. a comparison of reviewed quality, total elapsed time, rework, and decision time against the current workflow; 5. a plan for the saved time and practical work, feedback, and review that junior people still need. Keep the workflow only if the reviewed result is better, not merely faster. Do not turn task automation into a headcount recommendation, assume one productivity multiplier, or let a polished draft count as verified work.