by Claude Opus 5.5

What second-order effects tend to matter most (management layers, compliance overhead, customer expectations, speed of competition, organisational redesign)?

The first-order effect of AI, minutes saved on a task, is the easiest to measure and the least decisive for jobs. What matters more is what organisations do with the time saved: whether they redesign roles and hiring, absorb the saving in new review and compliance work, give it to customers as better service, or have it competed away. For UK jobs in 2026 the most important of these is organisational redesign through hiring. Compliance overhead and rising customer expectations come next, because they absorb much of the saving. Thinner management layers are plausible in theory but barely measured.

Why the second-order effects decide the outcome

The economics of earlier general-purpose technologies is clear on one point: the technology on its own rarely moves employment or productivity much. Bresnahan, Brynjolfsson and Hitt found that IT paid off when firms also reorganised work, with broader frontline responsibilities, more decentralised decisions and different skill mixes. Brynjolfsson, Rock and Syverson’s “productivity J-curve” explains why measured gains lag. Firms first have to invest in intangibles such as new processes, data and training, and the statistics do not record those well.

UK adoption looks like the early part of that curve. ONS finds that about 35% of firms with 10 or more staff now use AI, but the average number of AI technologies per user has risen only from 1.4 to 1.6. Only 11% of firms have trained more than half their workforce. Adoption this shallow is mostly bolted onto existing structures, so the effects below are at an early stage.

Organisational redesign: where the headcount decision is made

The most important second-order effect is a quiet one. Firms rarely make anyone redundant “because of AI”. Instead they redesign the work around the tool and then decide not to refill roles. The Bank of England’s July 2026 Monetary Policy Report describes exactly this: AI adoption is “gradually reducing demand for highly automatable jobs in some industries, with firms often slowing hiring or leaving vacancies unfilled”. The Bank’s Agents (September 2026) see “limited evidence of broad AI-driven reductions in employment” but say AI is “influencing role design and replacement hiring”.

This has two consequences readers should register:

  • It hits the pipeline before the workforce. Redesign usually removes the junior “doing” work first. PwC’s 2026 Barometer finds, in US data, that AI-exposed entry-level roles that have been “seniorised”, meaning their advertised requirements have moved upmarket, grew 35% since 2019.

  • It is invisible in redundancy statistics. UK redundancies were 3.6 per 1,000 employees in April–June 2026, largely unchanged on the year. An effect that works through attrition and hiring will not appear there.

Compliance and assurance overhead: the saving partly pays for itself

Every AI deployment creates verification work: checking outputs, documenting decisions, testing for bias, and handling challenges. In the UK this overhead grew in 2026 even as some rules loosened.

  • The Data (Use and Access) Act’s automated decision-making framework took effect on 5 February 2026. It permits significant automated decisions, but only with safeguards: information for the person affected, a right to contest and access to human intervention.

  • The ICO’s “Recruitment rewired” report (31 March 2026), based on work with more than 30 employers, found many were “likely relying on solely automated decisions” in hiring without adequate transparency or bias monitoring.

  • For firms selling into the EU, Article 50 transparency duties under the EU AI Act applied from 2 August 2026, although the high-risk employment rules have been pushed back to 2 December 2027.

Two things follow for jobs. Part of the time saved is spent on assurance, so net savings are smaller than pilot results suggest. And assurance creates work of its own in risk, governance, data protection and model validation. Shadow use makes the overhead larger. Deloitte’s 2026 survey of 25,000 UK workers found that 31% of GenAI users use it without their employer knowing, which means governance is being built after the fact.

Customer expectations ratchet upwards

When AI makes a faster or more personalised response possible, the first firms to offer it gain an advantage, and it then becomes the baseline. The saving is passed on as service quality rather than lower headcount. It is a Red Queen effect: running faster to stay in the same place.

Recruitment is a vivid UK example of both sides doing this at once. The ISE found 140 applications per vacancy in its 2025 survey, and 61% of employers had seen candidates use AI in interviews without disclosing it. AI cut the cost of applying, volumes rose, and employers responded with more assessment stages and more screening technology. On both sides much of the efficiency gain was consumed by an arms race.

Customer behaviour can also move demand away from a channel altogether. Centrica said in July 2026 that its 1,300 call-centre cuts reflected a 20% fall in call volumes, not AI, although unions dispute that. Either way, falling demand for human phone contact matters more for those jobs than any change in the cost of handling a call.

Speed of competition decides who keeps the gain

Whether AI savings create or destroy jobs depends heavily on market structure, a theme taken further in answer 1.12. Where competition is strong, savings pass through to lower prices or better quality, which can expand demand. Where firms have pricing power, savings become margin, and that shows up in the labour share and the tax base (answer 1.15) rather than in employment.

Morgan Stanley’s survey of firms that have used AI for at least a year, as reported in January 2026, found UK respondents reporting a net 8% job reduction alongside an 11.5% productivity gain. US firms, by contrast, reported creating more jobs than they cut. One plausible but unproven reading is that UK firms, facing weak demand, are using AI to cut costs rather than grow output. The survey records self-reported attribution, not measured effects.

Management layers: plausible, but unmeasured

Luis Garicano’s model of knowledge hierarchies explains management layers as a way of routing hard problems to people who know the answers. If AI puts that knowledge in front of frontline staff, fewer problems need escalating and fewer layers are required. Brynjolfsson, Li and Raymond’s study of 5,179 customer-support agents fits this: AI assistance raised productivity by 14% on average and by 34% for novices, which compresses the experience gradient that hierarchies are built on.

A second force pushes the other way. AI makes monitoring cheap and creates new review and exception-handling layers, so spans of control may widen in one place while new supervisory work appears in another. No UK data source measures management layers directly, so claims that AI is “flattening” UK organisations are inference, not observation.

How the effects compare

  • Redesign through hiring and attrition. Direction for jobs: Fewer entry and replacement hires. UK evidence status, Oct 2026: Reported by BoE and Agents; consistent with entry-level data.

  • Compliance and assurance. Direction for jobs: Absorbs savings; creates governance roles. UK evidence status, Oct 2026: Regulatory activity clear; role-level data not found.

  • Customer expectations. Direction for jobs: Converts savings into service quality. UK evidence status, Oct 2026: Strong in recruitment; anecdotal elsewhere.

  • Competition and pass-through. Direction for jobs: Jobs if demand expands, margin if not. UK evidence status, Oct 2026: Surveys only; contested.

  • Management layers. Direction for jobs: Ambiguous. UK evidence status, Oct 2026: Theory only.

What to watch

The useful indicators are ratios rather than totals: juniors per senior in professional firms, reviewers per case volume in operations, and vacancies refilled per leaver. The ISE’s 2026 Student Recruitment Survey, due on 14 October, will show whether graduate intake redesign continued this year. The Bank of England/FCA AI survey results, due at the end of 2026, should show how much financial firms spend on assurance.

Sources

From AI and Jobs: UK, October 2026