by Claude Opus 5.5
Which interpersonal and organisational skills tend to rise in value as technical output becomes cheaper (stakeholder alignment, prioritisation, accountability, explaining trade-offs)?
When drafts, analyses and code become cheap, the scarce work is deciding what to produce, getting people to agree to it, and putting your name to the result. The skills that gain are problem framing, prioritisation, stakeholder alignment, explaining trade-offs, accountable judgement, reviewing others’ work (including a machine’s), and developing other people. They pay off most when combined with real domain or technical knowledge. On their own they are much weaker protection.
Why the bottleneck moves
When one input to a piece of work becomes cheap, value shifts to whatever is still scarce. Before generative AI, David Deming’s research on the US labour market found that jobs needing a lot of social interaction grew by nearly 12 percentage points as a share of employment between 1980 and 2012. Employment and wage growth was strongest in jobs needing both high maths skill and high social skill. AI pushes further in the same direction by making more technical output routine.
Two well-known studies show what that looks like at work. Brynjolfsson, Li and Raymond studied 5,179 customer-support agents. An AI assistant raised issues resolved per hour by 14% on average and by 34% for novice and low-skilled workers, with minimal effect for the most experienced. Dell’Acqua and colleagues’ field experiment with consultants found that GPT-4 raised quality by more than 40% on tasks inside the AI’s capability “frontier”. Outside it, consultants using AI were 19 percentage points less likely to get the right answer. Knowing which side of that line a task sits on is a matter of judgement, not technique.
UK data point the same way. PwC’s 2026 Jobs Barometer found that since 2012, job postings in the occupations least exposed to AI grew 2.27 times, while the most exposed quarter barely moved (0.98 times). PwC’s US data add that exposed entry-level roles are becoming more senior. Their content shifts towards deciding, coordinating and checking.
The skills, and what they look like in practice
1. Problem framing. AI will answer whatever question you give it, so the value lies in asking the right one. That means turning “sales are down, look into it” into a defined question, with a measure of success, known constraints and an agreed view of what evidence would change the decision. Practice: before starting any significant piece of work, write a three-sentence problem statement and get the person who asked to agree it.
2. Prioritisation and saying no. When anything can be drafted in an hour, the backlog grows faster than ever. What constrains an organisation is attention and coordination, not production. People who can rank work by value against coordination cost, and decline politely, become more valuable. Practice: keep a visible “not doing” list and explain why each item is on it.
3. Stakeholder alignment. When everyone can produce a persuasive deck, persuasion depends less on polish and more on legitimacy. Who was consulted? Whose concerns are reflected? Who could block this, and have they been heard? This is the slow work of mapping interests, testing proposals early and building coalitions. AI can draft the email, but it cannot be the person whom the finance director trusts.
4. Explaining trade-offs and uncertainty. Leaders now receive more analysis than they can take in. The person who can say “option A is cheaper but locks us in for three years; option B costs £40,000 more and keeps our choices open” saves everyone time. So does putting the same decision three ways: in outcome and risk terms for executives, in workflow terms for frontline teams, and in control terms for compliance. Being able to say “we’re about 70% confident, and here’s what would change that” without sounding evasive is part of the same skill.
5. Accountable judgement. Someone still has to sign off the audit opinion, the clinical decision, the credit policy, the release to production. Regulators are clear that responsibility stays with the professional. The GMC says doctors “are responsible for the decisions they take when using new technologies like AI”. The SRA tells law firms they “cannot delegate accountability to an IT team or external provider”. Being someone who will make the decision, record the reasoning and stand behind it is scarcer than it sounds. Many people are happy to produce material, but far fewer will own the outcome.
6. Reviewing and editorial judgement. As more first drafts come from machines, the ability to review them well becomes a core skill. That means spotting what is plausible but wrong, what is missing, and what is technically correct but unwise. It is a different skill from producing the work, and many organisations have not yet trained for it.
7. Developing other people. This is easy to overlook and may be the most important. The ISE’s 2026 development survey found 29% of large employers reporting rising performance problems among new hires, up from 12% in 2022. Meanwhile, the typical development budget had fallen 10%. As AI absorbs the routine tasks juniors used to learn on, someone has to redesign how they learn. Managers and senior staff who can coach judgement, explain why something is wrong rather than just fixing it, and create deliberate practice become valuable to any organisation that wants a future talent pipeline.
8. Trust and relationships with a history. AI can write an empathetic message, but it cannot carry the five years of reliability that make a client take your call. The value of a relationship comes from continuity and accountability, and those cannot be produced on demand.
What is overrated
“Soft skills” as a generic label is too vague to be useful. It is also misleading on its own. A strong communicator with nothing substantive to communicate is not protected. The combination is what holds value: the engineer who can explain trade-offs to the board, the accountant who can align operations and finance, the nurse who can redesign a pathway and bring colleagues along. Deming’s finding was about jobs needing both maths and social skill, not social skill alone. Treat these skills as multipliers on expertise, not substitutes for it.
How to build them and show them
These skills develop through repeated practice in real situations. Volunteer to:
run the cross-team meeting;
write the decision paper;
review a colleague’s AI-assisted draft;
mentor the new joiner;
own an incident review.
Record evidence as you go. A decision memo you wrote, a conflict you resolved, a trade-off you made explicit or a junior you brought on all make concrete interview stories. They are more convincing than claiming to be “a strong communicator”.
Bottom line
As production gets cheaper, value moves to deciding, aligning, owning and teaching. These are not new skills. They have always been what distinguished senior people. AI asks for them earlier in a career, and it penalises their absence more quickly, because cheap output without judgement just produces more material that someone else has to check.