Skip to content
Houtini.
Contact
Field study · live data, primary sources Live · YubHub

In twelve months, AI used well has become a load-bearing productivity source.

And that's no marketing claim - we're looking at the data and watching AI proliferate in real time. But don't worry, AI isn't coming for your job. It's coming to make your job so much easier, and the exciting bit is that means you, or your team, can concentrate on the creative work, the strategy and the planning.

This page is the longer reading. 15,438 live job ads from the YubHub index, joined to Anthropic's Economic Index and cross-checked against the LinkedIn Work Change Report and BCG's reshape analysis. We name what AI is doing inside specific jobs. We name what it isn't doing. Where the data conflicts with the headlines, we say so.

15,438
Live job ads in dataset
784
Employers tracked
21 wk
Observation window
4 May 2026
Last refresh
01 · The work
The grunt work AI does well, mapped to the jobs that have it.

The Anthropic Economic Index publishes, quarterly, a sample of what people are using Claude for, classified against O*NET's task statements and aggregated up to the occupation level. The number Anthropic calls observed exposure is the share of an occupation's tasks that show up in real Claude conversations. It is a measurement, not a forecast. Computer Programmers sit at 0.75. Customer Service Representatives at 0.70. Data Entry Keyers at 0.67. Most jobs sit at zero, because most jobs are not at a keyboard.

The matrix below joins that data to live YubHub listings - what employers are actually hiring for, joined to AEI on SOC code. Read the right-hand columns together. A high AEI score plus a high listing count means the same thing every time: AI is doing meaningful work inside the role, and employers are still hiring people to do that role. That is augmentation in the data, not displacement.

What work is Claude doing?
Anthropic Economic Index task data · YubHub live listing volume · /stats/occupation-ai-exposure
Live · YubHub

Anthropic sampled a week of Claude.ai conversations (5-12 February 2026) and tagged each one to an O*NET occupation. The score is the share of that occupation's tasks getting done on Claude right now, so it's pulled from real usage rather than forecast or theory. Showing six occupations our SME readers care about, sorted by exposure.

Occupation For What AI is doing in this work Live listings AEI score
Customer Service Representatives
SOC 43-4051
E-commerce
  • first-line message triage and structured response
  • order-status and refund queries
  • knowledge-base search and quote generation
151 0.70
Data Entry Keyers
SOC 43-9021
Operations
  • reading source documents and entering structured data
  • normalising messy inputs into sheet schemas
  • cross-checking entries against source
2 0.67
Market Research Analysts
SOC 13-1161
Marketing
  • survey-data synthesis across mixed sources
  • keyword and topic-cluster research
  • long-form competitor and category briefs
134 0.65
Marketing Managers
SOC 11-2021
Marketing
  • first-draft campaign briefs and audience research
  • competitor read-outs and SERP analysis
  • performance-report assembly and slide drafting
503 0.32
Software Developers
SOC 15-1252
Tech build
  • first-pass code, tests and documentation
  • diff review and refactor proposals
  • Liquid / theme work, plugins, database queries
1,978 0.29
Operations / Business Analysts
SOC 13-1199
Operations
  • SQL extraction and dashboard interpretation
  • spreadsheet reformatting and data clean-up
  • weekly-readout drafting from raw figures
50 0.18

AEI score is the share of occupational tasks observed in real Claude usage (Anthropic Economic Index, March 2026 release). Live listings come from YubHub's job-feed index, joined to the AEI dataset on SOC code. Tasks shown are curated from the AEI per-task breakdown into buyer-facing language. Source.

Live skill volume
The most-listed skills cluster in the augmentation band.

Python, Communication, SQL, Project Management - the working skill set sits in the middle. AI is involved, the work continues, the skill is still listed in volume.

Skills where AI shows up in the work, augmentation at scale
50 skills · Anthropic Economic Index methodology · /stats/skill-ai-exposure
Live
Amplified · humans + AI Native · work the model does directly
0.0 0.1 0.2 0.3 0.4 0.5 50 100 250 500 1,000 2,000 AI exposure (Anthropic Economic Index, March 2026) listings (log) Python Communication Data Analysis Project Management SQL Java Machine Learning Problem-solving Artificial Intelligence Leadership C++ Communication Skills Kubernetes Go JavaScript Distributed Systems Customer Service AWS Teamwork Sales Excel Analytical Skills Data Science Collaboration Product Management TypeScript Large Language Models
02 · What stays

Strategy. Judgement. Knowing which result to act on.

The work that stays with people is the work people were hired to do. Strategy. Judgement. Domain context. Knowing which numbers in the dashboard matter and which are noise. Stakeholder management and the soft work of getting decisions through. The integration of multiple sources of information into a coherent decision. Anything that requires being in the room.

This is not a sentimental claim. It is what the AEI data shows when you read it carefully. Computer Programmers are 75% covered AND being hired in volume. Software Developers sit at 0.29 exposure, not 0.75 - because the work senior developers do is the integration work, not the first-pass code, and the integration work is not what AI is observed doing. The same shape repeats across Marketing Managers, Operations Analysts, and Market Research Analysts.

Klarna learned this in 2024-2025. They replaced 700 customer-service agents with an OpenAI-backed system in early 2024, then quietly walked it back in mid-2025 when CSAT collapsed. Triage and structured response, AI handles. The actual customer relationship - the part the role exists for - needs a person.

Where AI is doing the most work, and hiring is still happening
Anthropic Economic Index methodology · /stats/occupation-ai-exposure
Live
Low (<0.25) Moderate (0.25-0.5) High (>=0.5)
0.0 0.2 0.4 0.6 0.8 Customer Service Representatives 0.70 · 151 Data Scientists 0.46 · 578 Marketing Managers 0.32 · 503 Computer Occupations, All Other 0.31 · 145 Software Developers 0.29 · 1,978 Computer Systems Analysts 0.28 · 136 Management Analysts 0.24 · 140 Computer Network Architects 0.20 · 387 Advertising and Promotions Managers 0.17 · 535 Computer and Information Systems Managers 0.16 · 146 Computer Hardware Engineers 0.14 · 321 Producers and Directors 0.09 · 210 Managers, All Other 0.07 · 149 Sales Managers 0.04 · 350 Architectural and Engineering Managers 0.03 · 166 AI exposure score (Anthropic Economic Index, March 2026)
03 · What this looks like in your team

The grunt work that goes, the strategy that's now reachable.

Drawn from our own client work and the AEI task data - what changes for three of the most common buyer surfaces. Not "you must adopt or perish." This is what's becoming possible.

Marketing leader

Research, briefs and reports stop being the bottleneck.

Competitor reads, SERP analysis, keyword research, first-draft briefs, weekly performance read-outs. The work that used to take a senior person half a Tuesday now takes thirty minutes and a review pass. The senior person spends Tuesday on the strategy work the role exists for.

E-commerce / operations

Shopify, analytics and the data layer become reachable.

Theme updates, customised Liquid extensions, plugins, database queries - managed via the Shopify CLI inside Claude Code, end-to-end without leaving the editor. SQL extraction from GA4, GSC and the order-database, joined and explained, in minutes rather than half-days. First-line customer messages handled at the API layer; the human handles the relationship.

Owner / operator

Landing pages and the small-build work stop needing an agency.

Landing pages and small site builds - from scratch with the right skills library, or from a Figma design via a design-to-code path. Charts and diagrams generated through the Gemini MCP in the same session. Content research run as a multi-tool flow that reads, synthesises and drafts in one pass. The kind of work that used to need a brief and a vendor.

04 · Where I keep landing

The SME of twenty or thirty people is the most interesting reader on this page.

Most of the agentic-AI conversation is set in the language of multi-billion-dollar companies. Stanford payroll panels. BCG enterprise surveys. Foundation-lab usage data. That is not the most interesting reader. The most interesting reader is the SME of twenty or thirty people. For that team, this is the cheapest productivity uplift available right now.

And, crucially: it does not require adapting to someone else's way of working. Adopting a SaaS reshapes your procedures around the SaaS. An agentic framework reshapes around how you already work today. That distinction is large.

The SME read of all this is not "automate the routine work and fire half the team." It is the opposite. Free the people you already have - the ones whose judgement and context you spent years building - from spreadsheet reformatting and slide drafting and meeting prep, and let them solve new problems faster. Strategic action by people who already know the business is worth more than headcount reduction.

That is my read. The data on the rest of this page is primary-source and not in dispute. This section is opinion.

05 · The macro picture

The market is rotating, not retreating.

LinkedIn's Economic Graph team report 1.3 million new AI-related roles in two years and 600,000 new AI-enabled data-centre jobs. AI Engineer is one of the fastest-growing titles on the platform for three years running. Roles requiring AI literacy in the US are up 70% year on year.

Our YubHub feed shows the same shape on a different sample. Forward-Deployed Engineer, Applied AI Engineer, Deployment Strategist, Agent Reliability Engineer - titles that did not exist as named categories three years ago, now appearing several times a week. BCG's microeconomic model puts 50-55% of US jobs as reshaped over two to three years, with a far smaller share replaced. The consultancy register is what it is, but the directional read lines up with the live data.

The volume curve is steady. Sectors leaning into AI are accelerating their hiring more often than not. The market is alive and reallocating.

Weekly job-ad volume, twenty-week window
15,275 ads · 20 weeks · /stats/hiring-velocity
Live
0 2,000 4,000 6,000 peak 6,287 2025-49 2026-00 2026-03 2026-06 2026-09 2026-13 2026-17
Industry momentum, top four sectors
11 accelerating · 5 stable · 1 decelerating · /stats/industry-momentum
Live
  • Technology
    10,077 ads · twelve-week window
    accelerating +305%
  • Automotive
    1,709 ads · twelve-week window
    accelerating +573%
  • Finance
    1,028 ads · twelve-week window
    accelerating +632%
  • Manufacturing
    335 ads · twelve-week window
    accelerating +521%
06 · A distinction that matters

Exposure is not substitution.

The Anthropic measure tracks what AI is observed doing, not what employers have decided to do with the time it saves. The two are not the same and the distance between them is where most of the actual employment story lives.

You can see this in the data already on this page. Software Developers - the highest-listing occupation in our set - sit at moderate exposure (0.29) and are being hired in volume. Customer Service Representatives sit at the very top of the AEI scale (0.70) and are also still being hired. The companies doing the work have decided that the model is a tool for these roles, not a replacement for them. Anthropic's March 2026 release says so explicitly: there is no systematic increase in unemployment for highly exposed workers since late 2022.

Klarna's customer-service reversal is the worked example. Replace the role, lose the relationship; CSAT collapses; the role comes back, augmented. The pattern, not the headline, is what to plan around.

Advertised salary by sector, IQR with median
12 sectors with n >= 8 · /stats/salary-by-sector
Live
$0 $200k $400k $600k $800k Product management n=23 Legal n=123 Engineering n=4404 Design n=116 Customer success n=13 Marketing n=254 Sales n=612 Finance n=434 It n=68 Hr n=65 Security n=12 Operations n=310 advertised salary (USD, where stated)
07 · The honest counter

The entry-level question - taken seriously, not led with.

Abigail Marks at Newcastle argued in The Conversation that the fear of AI replacement is doing measurable damage right now to wellbeing and to the willingness of workers to plan, even where the technology has not replaced anyone. She is right about the distribution of harm. For workers in tightly-monitored repetitive roles - call centres, data entry - the fear is a daily presence whether or not the displacement is.

The Stanford Digital Economy Lab found a 16% relative decline in employment for workers aged 22-25 in the most AI-exposed occupations. Anthropic's own data finds a 14% drop in young-worker hiring into the same roles, "barely statistically significant" by their description. The Stanford February 2026 follow-up tightens the firm-time fixed-effects controls and rejects the rate-cycle alternative explanation, while moving the start of the AI-driven decline to 2024 rather than 2022.

The Yale Budget Lab and the Economic Innovation Group push back on the magnitude. They argue that the rate cycle, post-pandemic rebalancing, and the cooling labour market explain more of the entry-level decline than AI does. The conservative read across all of it: a real, AI-driven effect, smaller than headlines suggest, concentrated post-2024, confounded by macro factors that do not eliminate it.

This sits underneath the productivity story. Both are true. We do not lead with the entry-level finding because most of our reader's day is governed by what AI is doing right now in the work they manage, not by what is happening to the bottom of an industry's hiring funnel - but if you are a hiring leader, this is on you. Automate the routine work and architect the apprenticeship at the same time. The senior pipeline you will need in three years comes from the juniors you currently don't want to hire.

The pipeline paradox, what gets hired versus what gets trained
14,551 ads · senior-to-entry ratio 4.0 · /stats/experience-levels
Live
Current advertised distribution
  • Executive 482
  • Staff 909
  • Senior 7,789
  • Mid-level 3,442
  • Entry-level 1,929
What this snapshot can't tell you

The chart on the left is a single moment. It doesn't show movement.

For movement, see the Stanford Digital Economy Lab analysis of ADP payroll data, the "Canaries in the Coal Mine" working paper from Brynjolfsson, Chandar and Chen.

A 13 to 20 per cent decline in employment for 22-to-25-year-olds in AI-exposed roles such as software development and customer service.
Source · Stanford Digital Economy Lab, 2025
08 · What to do

Three things you can do this quarter - without us.

Pick one. The bar is low and the kit is mostly free. None of the productivity claim above lands until you have spent a few hours actually doing the work.

For the build-side reader

Get a working Claude Code setup running.

An afternoon. We have a getting-started write-up that covers the install, the first useful workflows, and the gotchas that catch people on day one.

Read the guide ↗

For the analyst / operator

Install your first MCP in Claude Desktop.

An MCP server is the cheapest way to give Claude access to your own data. Twenty minutes of setup; permanent productivity dividend. The walk-through covers our own MCP catalogue too.

Read the guide ↗

For the beginner

Try the Gemini MCP for charts and diagrams.

No code. Generate a chart, a diagram or an image inline in a Claude conversation. Useful in itself; also the gentlest possible introduction to what an MCP server actually does for you.

See Gemini MCP and the rest ↗

If you do work through one of those and end up wanting help building the agent layer around how your team actually works - instead of bending your team around someone else's SaaS - that is what we are useful for. Not before.

Reading list · primary sources cited above

If you want to argue with the readings, start here.