AI and the Future of Work in 2026

When Klarna’s CEO Sebastian Siemiatkowski told Bloomberg in mid-2025 that the fintech was hiring human customer support agents again — after eighteen months of arguing its OpenAI-powered assistant had quietly replaced seven hundred of them — the AI-replaces-everyone narrative cracked publicly for the first time. He admitted that “cost unfortunately seems to have been a too predominant evaluation factor” and that quality dropped. By 2026, the future-of-work conversation has matured past the cartoon. Some categories of work are getting displaced quickly. Others are being meaningfully augmented. A third group is barely touched yet. And a new tier of AI-adjacent jobs is materialising in real headcount numbers.

This is the picture as the second half of 2026 begins — what we are actually seeing inside enterprise clients, and what the named players have publicly committed to.

The displacement is real, but narrower than the headlines#

A few categories are demonstrably losing roles, fast.

Tier-one customer service and call centres#

Tier-one customer service is the cleanest example. Repetitive password resets, order status lookups, refund processing, basic policy questions — agentic systems handle these at a fraction of the cost. Marc Benioff has stated publicly that Agentforce is now handling somewhere between thirty and fifty percent of Salesforce’s internal workload, and Salesforce cut more than four thousand roles in 2025. Indian BPO operators, which built their growth on exactly this kind of voice and chat work, have begun re-pricing contracts and reshaping headcount. Cognizant and Infosys have both publicly tied a chunk of their workforce planning to AI-driven productivity gains.

Content moderation, basic copywriting, simple translation#

Content moderation queues at large platforms have shrunk meaningfully. Basic SEO copy, product descriptions, simple translation jobs — these have collapsed in price on freelance marketplaces. The work is being done in part by AI tools, in part by smaller human teams supervising AI output.

Data entry, basic accounting, low-end paralegal#

Invoice coding, receipt reconciliation, document indexing, contract clause extraction — these workflows have been quietly automated inside Big Four firms and large in-house finance teams. None of it is announced as a layoff line item; it shows up as “we didn’t backfill that team.”

Repetitive coding tasks#

Junior coding work that looks like “translate this Excel into a script,” “write the CRUD endpoints,” “draft the unit tests” has compressed. GitHub Copilot, Cursor, and Claude Code now do a respectable first pass. This is not the same as “AI replaced engineers” — see the augmentation section — but graduate hiring at several large tech employers has visibly slowed.

Where AI is augmenting, not replacing#

Bigger categories are getting faster, not smaller.

Senior engineering and architecture#

Senior engineers are shipping more, not being shipped out. The pattern we see across our clients in finance, healthcare and logistics is the same: senior engineers paired with coding agents close work in roughly half the time. The roles being squeezed are the most junior. The roles being elevated are the staff and principal engineers who can review, design, and integrate.

Research, analysis, and consulting#

McKinsey, BCG, Bain, and the Big Four have all integrated internal AI research assistants. Consultants are not being replaced; the analyst pyramid is being thinned at the bottom. The same is happening in equity research, legal research, and policy analysis.

Healthcare clinicians#

Ambient scribing tools — Abridge, Nuance DAX, Suki — are now in routine use at major US health systems. Clinicians are seeing more patients per day, not fewer clinicians being hired. Radiology, pathology, and dermatology continue to see AI co-read deployments, with the human clinician retaining final read.

Design and creative direction#

Tools like Figma’s AI features, Adobe Firefly, and Runway have raised the floor on production but not removed the designer. The work has shifted toward direction, taste, and brand integrity — judging and editing AI output rather than producing every pixel by hand.

Stylised desk with a coffee cup, headset, and folded paper chat bubble representing AI augmentation in customer support

The big-tech headcount story is more nuanced than “AI took our jobs”#

Microsoft has reduced headcount by more than fifteen thousand roles across 2025, including a roughly six-thousand-role round announced in May and a roughly nine-thousand-role round announced in July. Satya Nadella has publicly framed this as a reshaping toward AI-related work rather than a pure cost cut, and the company has continued recruiting aggressively into Copilot, Azure AI, and applied research teams.

IBM’s CEO Arvind Krishna has stated that several hundred HR roles have been replaced by AI systems, and IBM is making cuts described as “a low single-digit percentage” of its roughly 270,000 global workforce while pouring hiring into AI consulting and software. The framing inside the company is reshape, not reduce.

Cognizant and Infosys have both publicly signalled flatter graduate intake and productivity-led contract repricing. Indian IT services firms whose pricing model depended on linear headcount-to-revenue scaling are restructuring; the firms that win in this transition are the ones that move to outcome-based contracts.

The honest read: AI is not a unilateral cause of any single layoff round. It is reshaping the marginal hiring decision — should we hire ten more juniors or one senior plus tooling — and over time those marginal decisions add up.

Agentic AI’s actual ROI track record#

We need to be plain about this. Agentic AI has shipped real value in narrow, well-bounded workflows: customer-service deflection, sales prospecting, internal IT support, finance close support, marketing operations. It has consistently struggled when the workflow is open-ended, the stakes are high, and the supervision is thin.

Klarna’s reversal is the cleanest cautionary tale. So is the broader observation that many enterprise AI agent pilots in 2024 and 2025 quietly failed to graduate to production because measurement was thin and the agents either hallucinated, looped, or required so much human review that the cost gain disappeared. The pattern we recommend — and the pattern we see working — is assist, then automate: ship the AI as an assistant that a human approves first, measure for months, then progressively raise the autonomy ceiling for the narrow subset of tasks where the model has earned trust.

The four-day-week experiments now have data#

The UK’s 2023 trial with sixty-one companies and roughly three thousand workers ended with ninety-two percent of firms planning to keep the four-day week, and by 2025 more than two-and-a-half million UK workers reported working a four-day schedule. Iceland’s reduction in standard hours has held for nearly half a decade; productivity stayed flat or improved while reported wellbeing rose.

The interesting twist for 2026 is that several firms running four-day-week pilots have explicitly cited AI tooling as the enabling layer. RocketAir and other creative agencies have published case studies. Microsoft Japan’s earlier “Work-Life Choice” experiment, which paired shorter weeks with productivity tooling, is being revisited internally as Copilot adoption matures. The four-day week is not a universal answer, but it is no longer a fringe experiment.

Two stylised chairs side by side, one occupied by a paper-folded human silhouette and one by a circuit-pattern silhouette, representing human-AI co-working

What new roles are actually emerging#

Inside the enterprise hiring pipelines we work with, four role families are visibly growing.

AI ops and AI platform engineering#

The team that runs the model serving stack, the RAG indices, the eval harnesses, the cost and latency dashboards. This role barely existed in 2022. By 2026 it is a named function at most enterprises with serious AI investment.

Prompt and context engineering — pivoting#

The pure “prompt engineer” job posted in 2023 has mostly disappeared as a standalone role. The skill has been absorbed into product, engineering, and data science roles. Where it survives as a dedicated job, it has matured into something closer to “applied AI engineer” — owning prompts plus evals plus retrieval plus model selection plus cost.

AI risk, policy, and governance#

Banks, hospitals, insurers, and regulated tech firms are hiring AI-specific risk and compliance leads. The remit covers EU AI Act, US state laws, NIST AI RMF, internal model risk management, and vendor due diligence. We are seeing this as a director-level hire at most large enterprises.

Agent supervisors and human-in-the-loop reviewers#

Where agents are deployed in customer-facing or high-stakes workflows, the role of reviewing, correcting, and re-training the agent is becoming a real job. This often sits inside operations rather than engineering — it is the closest thing the 2026 job market has produced to a “new collar” role at scale.

What graduate hiring actually looks like in 2026#

The hardest conversation inside enterprise HR right now is about graduate intake. Several large tech employers — including some Indian IT services majors — have publicly trimmed graduate hiring numbers relative to peak years. The reasoning is rarely “AI replaces graduates.” It is closer to “AI has compressed the bottom of the work pyramid, so we need fewer entry-level seats to produce the same throughput.” The risk this creates is generational: if you do not hire and train juniors today, you do not have seniors in four years. The leadership teams we work with that have thought this through are explicitly protecting graduate intake at a slightly reduced number, redesigning the first eighteen months of the career path around AI-tool fluency, and treating the apprenticeship layer as a strategic investment rather than a cost line.

What this means for workforce strategy through 2027#

Three takeaways for senior leaders planning the next eighteen months.

First, do not run a single sweeping AI-replacement programme. The companies that have done that — Klarna being the public example — have walked it back. Run it function by function with measurement.

Second, invest in the augmentation pattern at the senior end of your workforce before automating at the junior end. Senior staff with good tooling produce more, with less risk, than a thin senior layer trying to supervise agents.

Third, hire deliberately for the four new role families above. They do not appear in classical org charts; you need to write them in.

Where pdpspectra fits#

We help enterprises ship AI to production without the reversal stories. That means evaluation harnesses, agent supervision patterns, cost and latency observability, and an honest sequencing of which workflows belong on humans, which on agents, and which on agents-with-supervisors. Our business automation practice covers exactly this terrain: scoping the workflow, mapping the human-in-the-loop checkpoints, and shipping the evals that catch regression before customers do.

If you are thinking through your 2026–2027 AI workforce plan and want a sober second opinion before you commit to a target operating model, we are happy to walk it through with you. The reversal stories cost a lot more than the planning conversation.