Revision note, October 5, 2026: Punctuation was normalized. Facts, sources, and editorial conclusions are unchanged.
A job title describes only part of the work. Before choosing training or preparing for a move, start with actual tasks: what you now do, what others expect and what you need to learn. This issue separates updating your current role from preparing for an adjacent one. It does not promise that everyone can stay put.
What actually changed
According to Revelio Labs, its October 1 update attributes 90% of measured year-over-year change in work activities in the US to changes within occupations; 10% comes from changes in the occupational mix. This decomposes an activity-dissimilarity index based on professional online profiles. It is not the share of workers affected by AI or jobs automated. The observations do not isolate AI's causal effect. They show why an occupation's name can conceal changing content, not what will happen to an individual worker.
A Federal Reserve research note, published September 30, finds broad AI skill requirements in 11% of online US manufacturing job advertisements, against 8% economy-wide. Requirements specifically for generative AI remain below 1% in manufacturing. The series ends in July and covers the industry, not just factory operators. The broad category includes machine learning and generative tools. Advertisements capture recruiting activity, not a count of open positions, completed hires or returns on training. The author's findings are not Federal Reserve policy.
Choose a next step for specific work, not training in “AI in general”: develop a missing skill within your role, or investigate a route into an adjacent occupation.
That is our practical interpretation, not the studies' forecast. Compare your work with employers' requirements. If a task remains useful but its methods change, test the missing skill on that task. If considering another role, separately check demand, credentials and acceptable pay.
Briefs: the week's signals
01. Running a business means reconciling different sources
According to Meta, Muse for Small Business adds connections to Shopify, Stripe, QuickBooks and other tools. Suggested uses include monthly financial reviews and unusual-expense flags. The company says publishing, sending and spending require approval; initial availability covers the US and Canada. Its examples do not establish reconciliation accuracy or time savings.
If combining business records, test a closed month. Do sales, payments and accounting entries agree under the same period and return rules? Choose a specific reconciliation gap to learn about, not an abstract course in running a business with AI.02. Ask about employees' experiences before judging adoption
According to Anthropic, its interviews about AI experiences and expectations run from September 29 to October 6. Publishing a full interview is optional. This is data collection, not findings. Claude users do not represent the population, and those agreeing to publication further select themselves. Removing names does not guarantee anonymity; copies retained by others cannot be recalled.
When discussing team training, examine a successful and an unsuccessful use of AI. What changed in the task, and what got in the way? Public research participation is a separate, voluntary choice; leave out confidential details.03. An analyst needs a measure's meaning, not just its value
According to Microsoft, Fabric IQ in Copilot Chat and Cowork uses Power BI semantic models: agreed business definitions and existing access controls. This capability is generally available; other announced components remain in preview. Connecting a data model does not establish that every answer is correct.
If preparing management reports, compare an assistant's answer about a familiar measure with the approved report. Check the definition, period and filters. The skill gap may be explaining the measure, not writing a longer prompt.04. Finance roles are expanding beyond budget scrutiny
According to IBM, 62% of 1,500 surveyed CFOs said their role had expanded into technology or AI strategy. Only 6% described their finance function as ready for transformation. The survey, conducted with Oxford Economics, ran from February to April. Self-reports and associations establish neither causation nor investment returns.
If your finance role is expanding, identify the new decision you contribute to: developing scenarios, setting investment priorities or interpreting data. Name the skill and a work example. Familiarity with a tool does not replace contributing to that decision.05. Updating a role does not rule out changing occupations
According to McKinsey Global Institute, its base scenario has about 11 million US workers potentially needing to change occupations by 2035, with a range of 6 to 16 million under other assumptions. This is a scenario, not observed layoffs. It includes AI, other automation, economic and demographic changes. Routes are assessed through demand, skill overlap, wage preservation and credentials.
If considering an adjacent role, check current vacancies, mandatory qualifications, preparation time and possible income loss. Establish whether the route is accessible locally. An available course does not establish an available job.06. Engineering work is shifting between writing and review
According to Terminal, 72% of 1,851 surveyed engineers say bottlenecks have shifted to review, testing and integration; 70% report spending more time reviewing than writing code. These are self-reports, not measured productivity gains. The sample is concentrated in particular countries, including Colombia and Mexico. This updates our earlier issue on workflow speed, rather than presenting the idea as new.
If your work has shifted toward review, measure waiting, corrections and the share accepted. Choose an exercise in testing or error analysis instead of accelerating generation further. Applying this elsewhere is a local hypothesis, beyond the engineering survey.07. Deployment skills include handing work to another team
According to BearingPoint, only 13% of organizations scaled AI fully in line with the original business case. Its study surveyed 1,050 leaders in 13 countries. Reports of financial effects concern the smaller subset with implemented AI. These are self-reports, not independently audited returns.
If you can build a pilot, test your next skill: handing it to an adjacent team. Provide instructions and result criteria, then observe questions and missing knowledge. Unrecorded help from the creator reveals what participants still need to learn.08. Data work includes checking relationships between tables
According to Google Cloud, Data Agent Kit is generally available, with BigQuery Graph, Bigtable and Spark additions. Its graph-building tool maps tables into nodes and edges, checks relationships against the data and proposes a plan before creation. These are claimed functions, not independently tested accuracy. The kit is free; the cloud services it uses are billed.
For data work, try relationships between products or suppliers and check the keys against known cases. Can you explain what each relationship means? Writing a query and understanding the data's structure are different parts of the job.09. Product selection starts with the shopper's task
According to OpenAI, a new Safeway experience turns a recipe, photograph or list into product selections and a cart. Customers go to Safeway for checkout. Other brands are planned; independently verified sales or quality results are not provided.
If working in retail, test the route from a shopping need to a cart: assortment, pack sizes, substitutions and final terms. Identify the part needing your professional judgment. A persuasive recommendation does not establish an appropriate selection.10. Agent users expect their roles to become harder to define
According to HP, 47% of surveyed knowledge workers using agents believe their role will become harder to define, against 39% using generative AI alone. Its survey covered 19,506 office workers across different groups in 15 countries, from April to May. These are expectations and group comparisons, not an established effect of agents.
If your role feels unclear, discuss what changed with your manager: tasks, expected results or skill requirements. Agree an observable expectation using a concrete work example before selecting training.11. Deployment training should finish with a working project
According to Anthropic, Claude Frontier Academy combines training, assessment and an organizational project. A 12-week practicum follows the in-person stage. Organizations nominate participants; this is not an open course for anyone. Announced investment and graduate targets concern the future. The program does not establish financial effects or replace the employer's assessment.
Before sending someone for deployment training, choose a process and the person accepting the result. Require a pilot with exceptions examined and a team handover. A certificate confirms training, not a change in your work.12. Incoming messages can be prepared for further handling
According to Anthropic, Barclays' Global Markets platform uses Claude to classify, enrich and route approximately 120,000 emails daily. The expanded-partnership announcement describes an operating workflow. Message volume does not establish routing quality or savings.
If handling incoming requests, examine past cases: which fields determine the right destination, what is missing and when clarification is needed? Preparing a request for handling may matter more than the speed of the first reply.What became cheaper, and what became more valuable
These sources establish neither a general price of work nor that every skill has become cheaper. Our conclusion is narrower: access to a tool does not tell you which part of your role to learn. Start with a required result and the gap between your current method and the new requirement.
Test learning on a task: can you now do something previously out of reach, explain your decision and use it at work? A feature matters when it closes that gap. The number of tools you know does not measure it.
Verified figures and primary sources
According to Revelio Labs, the comparison concerns year-over-year change in an activity-composition index derived from professional profiles. It does not count people keeping or losing jobs.
The Federal Reserve note observes online job advertisements; its skill-requirement series ends in July. Publication this week does not mean the measured changes happened this week.
How to read this
Profiles, advertisements, surveys and scenarios answer different questions. A requirement is not a hire; an expectation is not an observed change. An announced feature does not establish usefulness. The sources suggest what to investigate; your own role still needs examining.
The week's contrast
Changing tasks within an occupation and moving between occupations can coexist. Neither updating a skill nor immediate retraining is the answer for everyone. Check the requirements of current work or the accessibility of a new role. Neither a job title nor an attractive course settles that decision.
People: work and accountability
Start with what you actually did, not an idealized job description. A new task may need subject knowledge, data checking or an explanation; tool skills are only one possibility. Agree what success means with the person using your work, so training goes beyond learning an interface.
Business: decisions, economics and risk
Separate development within a role from preparation for another. The former needs a work example; the latter also needs demand and entry conditions checked. Revisit hiring and development requirements around tasks that actually changed. This is a proposed method, not evidence that adding a requirement increases income.
Trust: boundaries, verifiability and consequences
Do not turn an index into a promise to a person. Explain the observation, sample and remaining assumptions. A training recommendation is more credible when its task and assessment are clear. Career discussions should disclose unknown requirements, preparation time and possible income changes.
Working map for the week
These are our proposed local checks, not prescriptions from the studies.
| Decision | When it applies | First move | Boundary |
|---|---|---|---|
| Update part of your role | The title stays but tasks change | Compare actual tasks and expected results | The index does not predict your career. |
| Choose a missing skill | Your current work has a new requirement | Compare the task with relevant job advertisements | Broad AI skills extend beyond generative tools. |
| Investigate an adjacent role | You are considering a move | Check demand, credentials, preparation and pay | A scenario does not guarantee an accessible route. |
Continue in the Practicum
The Week Inventory helps reconstruct actual tasks from your calendar and correspondence. Use it as a starting list, not a forecast of occupational survival.
The One-Level Plan helps choose a testable change to a task over the coming quarter. It is a task plan, not an assessment of demand for another occupation.
One sensible next move
Compare a task from last week with a current work requirement. Name the missing skill and an example on which to test it. If considering a career move instead, first check real vacancies and entry conditions. Choose training after that investigation, not instead of it.
Confidence and sources
Publication dates and wording were checked against primary sources, linked beside the facts. US findings do not transfer automatically to another market. Surveys describe participants' responses, product publications contain supplier claims, and scenarios depend on assumptions. Practical moves and the working map are our proposals for limited local checks, not promises of job retention, higher pay or a successful transition.