Workday Cut Product Jobs, Then Opened an Office in Dubai
On this page 7 sections
AI-generated editorial illustration.
On September 29, Workday told investors it would reduce about 2.5% of its current workforce, with most of the affected positions in Product and Technology. Its Form 8-K also said the company would trim selected leased offices and keep hiring in strategic areas and locations. Two days later, Workday announced a new Dubai office, regional operations and expanded local leadership.
A regional customer hub and a product engineering group do different work. The public documents give no Dubai office headcount or list of affected product teams. They do not say AI caused the reduction. Meanwhile, Workday keeps adding AI products to a platform used by more than 11,500 organizations, according to its August earnings release.
A buyer waiting for an integration needs to know who will maintain it. A worker facing notice needs to know whether an open position is reachable. A shareholder has a charge estimate and revised GAAP margin guidance. The documents answer the investor’s immediate accounting question more precisely than the customer’s or worker’s practical one.
September 29 set a new staffing plan
Workday announced a plan, not a final payroll count. Its filing says “approximately 2.5%” of the current workforce, primarily in Product and Technology. It does not break the percentage into engineers, designers, researchers, managers and support staff. The fiscal 2026 annual report counted more than 21,000 employees on January 31, 2026. That older number cannot turn September’s percentage into a precise headcount.
Workday can put a range on the accounting charge even without publishing a headcount. It expects $65 million to $80 million in total: $40 million to $55 million of future cash spending for severance, benefits and related costs, about $10 million in noncash stock compensation and about $15 million in noncash impairment of leased office space. It expects $55 million to $70 million of the charges in its fiscal third quarter and $10 million in its fourth. These are estimates of costs, not money already saved. The employee actions are expected to be substantially complete by the first quarter of fiscal 2028, subject to local law and consultation. The property actions have a separate fourth-quarter fiscal 2027 target.
The same filing revised Workday’s GAAP operating margin outlook downward relative to its non-GAAP outlook because of those charges. Workday intends to exclude the reorganization charges from its non-GAAP measures. A reader who sees the $80 million upper bound beside a higher adjusted margin should keep the two accounting views separate. The charge records the cost of changing the organization. It does not measure the future quality of the software or the return from the change.
A second 2026 reorganization has a different center of gravity. Workday’s February filing projected a reduction of about 2%, primarily in Global Customer Operations, and $135 million of charges. The July quarterly report says that plan was substantially complete in the first quarter of fiscal 2027. It records $55 million in employee transition and related charges plus $80 million of space and asset impairments. September’s plan focuses mainly on another organization, with no disclosed AI-causation claim.
For a customer, the sequence poses a narrower service question. February changed Global Customer Operations. September targets Product and Technology. The first group helps a buyer use and maintain what has shipped; the second builds and changes the product. A reduction in one does not prove weakness in the other, and the filings offer no service-level data after either plan. But someone renewing an HR platform should ask for the handoff across those groups, especially when a new agent crosses application boundaries. A ticket that starts in support and ends with an engineering fix depends on both ends of the chain.
For an employee, the filing’s long completion window is more consequential than the rounded percentage. The workforce action can extend into the first quarter of fiscal 2028 because consultation and local law differ by place. It does not say how many people have already received notice, how many could move to strategic openings or whether the research and regional roles are open to affected staff. Those are separate outcomes, and the public percentage does not answer any of them.
Dubai adds a different kind of capacity
Workday’s October 1 announcement names Angelique de Vries-Schipperijn, president for EMEA, and Zakaria Haltout, a regional group vice president, at the UAE launch. It describes a new Dubai office, dedicated regional operations and more local leadership. It says the hub will engage customers and partners and attract local talent. It does not provide a budget, a hiring target or a list of roles.
The missing role list limits what can be concluded. An implementation may need local account leadership, sales engineering, specialists and support; the release confirms a regional presence but does not say which functions will be staffed or when. A new office also does not establish a net increase in floor space. September’s filing plans reductions in selected leased offices without naming them or reconciling total area and cost.
Workday published a company-backed UAE survey alongside the office news. It says employees reported saving an average 3.6 hours a week through AI and spending an average 3.4 hours correcting, clarifying or rewriting low-quality output. Those figures are useful as a warning against treating gross time saved as net capacity, but they are survey responses in a market Workday is entering. They do not estimate Workday’s own engineering productivity or the effect of its product-team cut. A regional launch can give buyers access to people without answering how the platform’s core software will be built and supported.
The same survey says 33% of respondents’ organizations use AI efficiency to handle more work without increasing headcount, 32% direct savings toward training and another 32% toward workload or flexibility. The release presents these as reported organizational choices, not as measured changes in jobs. The percentages need not be exclusive, and they cannot be mapped onto Workday’s own September plan. Their value here is to expose the variety of ways a company can use time savings. The announcement of a regional office is a bet on customer contact; it is not a published experiment showing which use of saved time will produce a better implementation.
The local-office move also creates a division of responsibility. A buyer may meet someone in Dubai about an HR workflow whose software components were developed elsewhere and whose infrastructure is hosted by third parties. Workday has not published that staffing map, so it would be wrong to identify any specific Dubai team as a support replacement. The buyer can still ask who has authority to diagnose the integration, who approves a fix and which support hours apply in its region. Those questions become more pressing when the vendor is changing both its product organization and market coverage in the same quarter.
Workday may be concentrating development on priority products while putting more people near regional buyers. The filing says strategic hiring will continue through fiscal 2027. Whether those hires replenish a function affected by the reduction or add an entirely different capability remains unknown.
Product spending rose before the cut
In its August 27 results, Workday reported $2.649 billion in second-quarter revenue, up 12.8% from a year earlier, and $2.471 billion in subscription revenue, up 13.9%. Its GAAP operating margin was 11.8%; its adjusted margin was 31.1%. Operating cash flow was $520 million, down from $616 million in the year-earlier quarter. Workday also spent $1.3 billion repurchasing shares and ended July with $3.403 billion in cash, equivalents and marketable securities.
The July 10-Q puts second-quarter product development expense at $747 million, versus $660 million a year earlier. Costs of subscription services rose from $370 million to $436 million. Workday attributes $33 million of that increase to third-party hosted infrastructure. These figures predate the September reorganization. Revenue grew, and spending on the product and its delivery grew too. The filing does not forecast a product-development dollar saving from the cuts.
CEO Aneel Bhusri said AI drove more than 25% of new annual contract value in the quarter and more than 5,500 customers used at least one of Workday’s organic agents. The company said the customer count had risen more than 35% from the prior quarter. These are Workday’s own commercial and usage measures. They do not state how many agents completed a sensitive HR or finance task correctly, how much customer labor was released or whether support demand changed. CFO Zane Rowe paired a 13% full-year subscription-growth outlook with an increased 31% non-GAAP operating-margin outlook, saying the company would prioritize its agent roadmap while seeking operating efficiencies.
Even within Workday’s financial reports, metrics answer different questions. The July 10-Q gives gross revenue retention of about 97%, which measures recurring revenue retained from a set of existing customers; it excludes expansion. New annual contract value, by contrast, is a sales measure. An organic-agent customer count says a customer uses at least one agent but not whether the agent serves one department or thousands of employees. An investor can track all three without pretending they are successive steps in one verified productivity calculation. A customer needs a different denominator: accepted work, correction effort and cost at its own deployment.
The quarter carried $1.3 billion of share repurchases, while operating cash flow fell from the year-earlier period. Workday has not said the reorganization funded the buyback. The figures show management making several capital choices at once: development, acquisitions, infrastructure, expansion, repurchases and payroll. They appear on different reporting clocks. None gives a dollar-for-dollar explanation of the September cuts.
Workday has to pay for development, hosted infrastructure, integration and regional growth while protecting adjusted margin. The public record shows those costs at company level. It does not allocate the September reorganization charge to a particular AI launch.
Acquisitions widen the integration load
Workday’s annual report lists four fiscal 2026 AI-related acquisitions: Flowise, Paradox, Sana and Pipedream. The acquired products bring agent building, conversational recruiting, enterprise learning and integration capabilities. Their purchase is not evidence that an internal Workday engineer’s position became redundant. Acquired teams and a legacy platform still need to agree on authentication, data permissions, release schedules and customer support.
The August release makes the work visible in pieces. Workday said Learning powered by Sana had become generally available. It described a Developer Agent and Agent Passport, a Financial Audit Agent that had become generally available, and integrations with AWS and Google Cloud. In August it also announced a dedicated AI Research team focused on agent memory, explainability, orchestration and efficiency. President of Product and Technology Gerrit Kazmaier described privacy, auditability and accuracy as problems the team would work on. A research arm and a released product have different delivery clocks; neither identifies which existing team lost September positions.
The engineering burden is not just to make an agent produce an answer. A recruiting agent can affect a candidate record. A finance agent can assemble an audit package. A learning tool can recommend content to an employee. Each task crosses permissions, integration, data freshness and the possibility of a wrong or incomplete result. Workday’s 10-Q warns that AI’s technical and legal environment could raise development costs or divert resources. That is a risk disclosure, not a report of a particular customer failure.
The four acquisitions make the delivery question more concrete. Paradox’s conversational applicant flow touches recruiting data and candidates; Sana’s learning experience touches employees and skills; Pipedream provides connectors; Flowise contributes agent-building tools. The annual report describes those capabilities, but it does not publish an integration dependency chart. A Workday customer assessing a new feature should be able to identify which component owns a record, whether an agent’s proposed action is reversible, and which product team or partner will take a fault report. That is a buyer’s proposed acceptance test, not a finding that any acquired component has failed it.
An HR leader buying the Sana learning experience and a finance leader buying the Financial Audit Agent will both encounter the Workday platform, but their acceptance tests should differ. For learning, the customer can ask whether a skills recommendation is explainable to the employee, reflects current job data and can be corrected. For an audit package, it can ask which source records were included, who approved the package and whether a reviewer can reproduce it. Workday has announced the products’ availability. It has not published a common outcome denominator that would let an outsider compare those two jobs as if they were identical AI tasks. The role of product and technology staff in those acceptance paths is precisely the missing organizational detail.
Integration also changes the size of a release. A correct answer in a demo is a narrow result. Keeping it correct after a permission change, connector update or regional policy difference requires regression tests and someone authorized to stop the release. Workday says Agent Passport will test and verify agents before production and monitor them after. That product description says nothing about the capacity of the teams carrying out reviews after September. A customer can request evidence from its own deployment. The layoff percentage cannot supply it.
Google’s DORA research studies AI-assisted software work as part of a delivery system; faster code creation alone is not a production result. NIST’s March 2026 report describes gaps in methods for monitoring deployed AI systems. Neither study audits Workday. For a customer, the relevant evidence after a team change is a tested release, observed live behavior and a way to correct failures.
DORA’s 2025 work drew on nearly 5,000 technology professionals and more than 100 hours of qualitative research. Its finding that AI amplifies strengths and dysfunctions in existing organizations is a warning against treating generated output as a staffing formula. NIST reached its monitoring observations through practitioner workshops and a literature review. It distinguishes pre-release tests from the harder job of seeing what a system does with changing inputs after deployment. The studies come from different settings, but both leave a practical point for the buyer: if a reorganization removes or moves people, identify the owner of the feedback loop before accepting a faster release schedule.
AI-generated editorial illustration. The paths separate product delivery from regional presence; they do not depict Workday’s actual organization.
A ledger for exits, hires and customer work
A buyer cannot obtain Workday’s internal staffing plan from an 8-K. A worker cannot infer an individual outcome from a company percentage. Still, both can use the same discipline: keep each claim beside its denominator, owner and verification date. The table below is an editorial assessment tool, not a leaked Workday plan or a claim that a named team is understaffed.
| Line to reconcile | Public signal by October 2 | Decision still needing evidence |
|---|---|---|
| Workforce exits | Approximately 2.5% of current staff, primarily Product and Technology, planned in the September 29 filing | Which functions, regions and customer commitments lose named owners? |
| Strategic hires | Workday says it plans to keep hiring in key areas and locations | Which roles open, are filled and restore or add capability? |
| Space | Selected leased-office reductions plus a newly announced Dubai office | What are net seats, costs and customer-facing coverage by region? |
| Product investment | $747 million second-quarter product development expense, before the new cut | Which roadmap, testing and integration responsibilities remain funded? |
| Commercial adoption | More than 5,500 customers using at least one organic agent, by Workday’s count | How many production workflows are accepted, corrected or escalated? |
| Customer service | Prior Global Customer Operations reduction substantially complete, according to the July 10-Q | Are support response, resolution and implementation quality stable by product? |
A procurement team renewing a Workday contract can ask for an owner for each integration, a supported release schedule, a rollback procedure and an escalation path for an agent decision that affects a person or a financial record. These are questions to put to a vendor, not facts Workday has failed to supply to every customer. The internal owner should also record what outcome would change the renewal decision: a tested deployment, successful case resolution, corrected data or a support threshold. Counting agents or features alone cannot answer that.
Employees facing an organizational change need a different version of the ledger. The filing says local law and consultation can affect timing. A useful internal record would distinguish notice, severance, open roles, redeployment eligibility and the handoff of a product obligation. Without that distinction, a company can count a strategic opening and an eliminated position in the same sentence while leaving the people and customers attached to each one invisible.
For a product manager, the ledger needs a time dimension as well. The September filing gives a forecast for charges and completion, while the August results describe an earlier quarter. Put a date beside every cost and service measure. If support response times deteriorate in November, a team would still need to establish whether the cause was staffing, a new agent rollout, an acquired connector, an incident or ordinary demand. The table prevents a tempting but unsound shortcut: drawing a line from two adjacent announcements straight to a product outcome.
The customer and employee versions should meet at the same handoff. When a product responsibility moves, the customer needs a named escalation route and a documented release owner; the affected employee needs to know whether knowledge transfer and redeployment are part of the plan. The public documents do not disclose those details. Asking for them is more concrete than either declaring the company efficient or assuming that a smaller team must deliver worse software.
The ledger should record a counter-signal as readily as a problem. If a feature ships on its promised date, a customer accepts it and support cases fall, that evidence would weaken concerns that the reorganization damaged delivery. If open positions remain unfilled, an integration slips and correction time grows, the opposite reading gains force. Those examples are proposed observations, not forecasts. They also keep the analysis fair to a company that has disclosed the scope and expected cost of a change but has not yet had a quarter in which its new structure can be evaluated.
Why the AI explanation remains unproven
Workday bought AI companies, reported agent use and then announced product and technology cuts. The filing attributes the reorganization to growth priorities without dividing the 2.5% reduction among AI, acquisition integration, management layers, location costs or other choices. It offers no causal account that would support an AI-replacement claim.
A large software company can remove some roles, hire specialists for new products and open near buyers without replacing every departing person with software. Workday’s annual report counted more than 21,000 people in 36 countries, and its September filing says it will continue hiring. That does not establish how employees or customers will fare. It would take post-change evidence: role-level openings and exits, product delivery, support quality and financial results after the charges settle.
Earlier percentages need care. The July 10-Q says a February 2025 plan ultimately reduced about 7.5% of the workforce, although the announcement had anticipated approximately 8.5%. It separately records the February 2026 plan at about 2%. Each used a workforce at a different point in time. Adding 7.5, 2 and 2.5 into a single cumulative share would be false. Applying September’s share to January’s headcount would also give false precision.
Workday’s UAE survey illustrates the same measurement problem on the demand side. Reported hours saved and hours spent correcting output can both be true for respondents. Neither becomes a net company productivity number until one knows whether the same workers answered both questions, what work changed and how the time was used. The announcement says 33% of surveyed organizations handle greater workload without adding headcount; that is a survey result, not evidence that Workday’s own reorganization saved the same amount of labor.
After the charge, a customer handoff
The September filing puts the next immediate financial effect in the fiscal third quarter, when Workday expects to recognize $55 million to $70 million of the new charge. Its earlier results provide a product-development and service-cost baseline. Later filings can show whether the cost mix changes. They cannot alone show whether a customer waiting for an integration got a stable release.
The next customer review will put a product owner, regional lead and support team around a specific ticket. If ownership changed, someone still has to test the fix and tell the customer what happened. Workday has disclosed the scale of its workforce cut and the location of a new office. The owner of that future ticket is absent from both announcements.