AI Has a Real Economy Problem? What UK, US and OECD Data Actually Show
UK payrolls fell and vacancies hit a 5-year low, but US productivity is accelerating. The gap is sectoral, not proof AI is failing.
Summary
Bob Elliott @BobEUnlimited says AI has a real economy problem with no productivity gains outside tech, citing UK data. The UK labour market is weak, but productivity is flat not collapsing, and US data shows the opposite trend. The evidence points to a classic General Purpose Technology lag, where tech leads and the rest follows years later, not a structural failure of AI.
The UK numbers that sparked the debate
UK data on Aug 18 triggered the discussion. Payrolled employees fell by 13,000 in July to 30.3 million on a provisional flash estimate, and by 78,000 year on year to June, according to the ONS Labour market overview, UK: August 2026. The same release was covered by The Guardian and Bloomberg.
Other signals from that release, highlighted by @elerianm:
- Payrolls -13k in July (flash), -78k YoY to June; vacancies 707,000 in May-July, down 6k on quarter — outside the pandemic, last at or below 707k was Sep-Nov 2014
- Regular pay ex-bonus +3.5% YoY overall, but private sector +2.8% (6-year low) vs public sector +6.1%
- Real regular pay only +0.5% (CPIH) / +0.7% (CPI)
- Unemployment 4.9% in Apr-Jun, unchanged on quarter but +0.2pp YoY — ONS warns Labour Force Survey estimates remain volatile and recommends focusing on PAYE Real Time Information for employee numbers
These are lagging indicators. They confirm demand has already softened, they do not predict whether AI will lift output per hour next year.
Productivity: flat outside tech, not collapsing
The ONS productivity flash shows a split by data source and by industry.
For Q2 2026, the ONS flash estimate for April to June 2026 reports output per hour up 0.7% year on year and output per worker up 1.4% on its preferred PAYE RTI-based method. On the Labour Force Survey-based method, output per hour was down 0.2% and output per worker up 0.4%.
For Q1 2026, the ONS flash estimate for January to March 2026 reported output per hour up 0.4% YoY on the LFS method and up 2.1% on the RTI method. Output per hour is only 2.3% to 4.6% above the 2019 average, depending on method, which is weak compared with the pre-2008 trend.
By industry in Q4 2025 versus the 2019 average, information and communication made the biggest positive contribution to productivity growth, driven by a large rise in gross value added with a smaller rise in hours. Human health and social work made the biggest negative contribution, with hours up sharply and output up only a little. That sectoral split is exactly the point Bob Elliott makes — tech is productive, the rest is not yet.
This pattern echoes earlier work on why AI does not kill software but crowns the data moat, where gains concentrate where data and distribution already exist.
Sources: ONS flash estimates Q2 2026, BLS TED Aug 11 2026, OECD Compendium 2026. As of Aug 19, 2026.
What the US and OECD show
The US is not following the UK path.
- The BLS Productivity Home Page reports nonfarm business productivity rose 1.4% annualized in Q2 2026, with unit labor costs up 1.3%. Output rose 1.7% while hours rose 0.3%.
- On a year-on-year basis, productivity was up 2.2% from Q2 2025 to Q2 2026, which is above the post-pandemic average.
The OECD picture is mixed but not collapsing. Across all OECD countries, labour productivity grew 1.2% in 2024, double the 2023 pace, but the median across members was only 0.4%, well below the 1.8% pre-crisis average. Growth is uneven and historically weak, but it is growth.
For investors, the US revival matters because it shows AI investment can coincide with measured productivity gains when output rises faster than hours, as seen in recent Cloudflare and Atlassian earnings where AI shows up in revenue and the SaaS split.
Structural or lagging?
Three layers to separate:
- Lagging labour market. Payrolls, vacancies and wage growth lag output. The UK weakness confirms a soft patch that began months ago, not a new AI-driven break. Small firms cite higher labour and operating costs as a reason to scale back hiring, per the ONS.
- Structural UK productivity. UK output per hour has grown only about 0.4% to 0.7% YoY on recent prints, in line with the 2009-2019 trend and far below pre-2008. That is structural and predates AI.
- AI-specific lag. A widely cited NBER survey of about 6,000 CEOs and executives found AI widely adopted but average productivity effect of only about 0.29% over three years and employment effect near zero, echoing Solow's paradox that you see computers everywhere but in the productivity statistics. Executives expected about 1.44% productivity gain over the next three years. Research on General Purpose Technologies finds the same pattern for steam, electricity and computers — intangible investments in process redesign and retraining take years before gains appear in GDP.
In short, the UK data supports "no sign outside tech yet," but the US 1.4% quarterly and 2.2% yearly prints argue against calling it a permanent AI failure.
What to watch
- ONS industry productivity for Q2 2026, due next quarter, to see if retail, manufacturing or finance start to show GVA rising faster than hours
- US Q3 productivity on Nov 6, and whether unit labor costs stay near 1% to 2% while output holds up
- UK vacancies and private sector pay in the September release, the cleanest real-time read on whether the labour market stabilizes or deteriorates further
- Company-level evidence outside tech, where capex and power constraints are now the binding factor, not model quality
Sources
- ONS Labour market overview, UK: August 2026 — payrolls, vacancies, earnings — ons.gov.uk
- ONS Productivity flash estimate Q1 2026 and Q2 2026 — output per hour and per worker — ons.gov.uk and ons.gov.uk
- BLS Productivity Home Page and TED Aug 11 2026 — US nonfarm productivity — bls.gov and bls.gov
- OECD Compendium of Productivity Indicators 2026 — 1.2% OECD growth in 2024 — oecd.org
- Bob Elliott @BobEUnlimited and Mohamed El-Erian @elerianm — X posts Aug 18 2026