In a special postbag edition, Financial Times readers have offered something more revealing than predictions: a ground-level account of how AI is already reshaping work, not in abstract terms but in the rhythms of daily tasks. The letters describe a shift that is uneven across industries and roles, arriving first where work is easiest to standardise, document, and measure. They also show that the biggest changes are often less about replacing people outright and more about reorganising responsibility—who checks, who decides, who signs off, and who absorbs the risk when outputs go wrong.
Taken together, the messages read like a map of early adoption. AI tools are appearing in inboxes, ticket queues, drafting workflows, customer support scripts, internal knowledge bases, and analytics pipelines. In many cases, the technology is being used as a “second set of hands” for routine labour: summarising long documents, generating first drafts, translating content, extracting key fields from messy data, or suggesting next steps based on patterns. But readers repeatedly emphasise that the real story is what happens after the first draft—how teams learn to trust, verify, and redesign processes around machine assistance.
One theme runs through the postbag: AI is impacting workflows, not just future possibilities. Several correspondents describe the moment when an organisation stops treating AI as an experiment and starts embedding it into the mechanics of work. That transition tends to be triggered by practical needs—speeding up turnaround times, reducing backlog, improving consistency, or lowering the cost of producing routine outputs. When AI is introduced with clear boundaries and measurable targets, adoption accelerates. When it is introduced as a vague promise, it stalls. Readers’ accounts suggest that the difference between success and failure is rarely the model itself; it is the operational discipline around it.
For some workers, the most immediate effect is time compression. Drafting that once took hours becomes a matter of minutes, at least for the first version. Customer-facing teams report faster responses and more consistent messaging, particularly when requests follow predictable patterns. Analysts describe quicker cycles for cleaning data, producing initial insights, and generating explanations that can be refined rather than built from scratch. Even in roles that require judgement, AI is often used to reduce the “blank page” problem: it helps teams start, then they apply expertise to correct, contextualise, and decide.
Yet the letters also show that speed can create new pressure. When AI makes output cheaper and faster, expectations rise. Readers describe a subtle shift: the work does not disappear, it moves. Instead of spending time producing the first version, employees spend more time reviewing, reconciling contradictions, and ensuring that the final product meets standards. In other words, AI can increase throughput while also increasing the volume of work that reaches the human stage. Some correspondents say they now handle more requests per day because the initial legwork is automated. Others worry that this creates a hidden intensification of labour—less time to think deeply, more time to supervise.
The postbag also highlights how automation tends to cluster around tasks that are repetitive, text-heavy, and rule-bound. That is where AI delivers the most visible gains. But readers’ experiences suggest that the technology’s influence extends beyond those tasks. Once AI is integrated into a workflow, it changes how teams structure information. People begin writing differently—more clearly, more consistently, with prompts and templates in mind. They also begin capturing data differently, because AI systems perform better when inputs are well organised. Several correspondents describe a “documentation effect”: the need to provide context, definitions, and examples so that AI outputs align with organisational norms. This can be beneficial, but it also forces teams to confront gaps in their own processes.
A second major theme is skills reshaping. Readers do not describe a simple replacement of human expertise with machine output. Instead, they describe a reweighting of competencies. Technical staff increasingly need to understand how to use tools effectively—how to craft prompts, how to interpret confidence signals, how to test outputs against known benchmarks, and how to iterate. Non-technical workers, meanwhile, are learning to treat AI outputs as raw material rather than finished answers. The skill is not only “using AI,” but knowing when not to use it, when to escalate, and how to validate.
Several letters point to a new kind of literacy: understanding outcomes and limitations. Workers are being asked to evaluate whether an AI-generated summary is faithful to the source, whether a recommendation reflects the actual constraints of a case, and whether a generated response introduces subtle errors. This is not merely a matter of spotting obvious mistakes. Readers describe the challenge of detecting plausible-sounding inaccuracies—errors that look reasonable but are wrong. As a result, training is shifting from generic “AI awareness” to practical verification habits: cross-checking facts, using citations where available, running consistency checks, and maintaining audit trails.
Employers, according to the postbag, are responding in different ways. Some organisations invest in structured training and governance, creating playbooks for acceptable use, escalation paths, and quality assurance procedures. Others roll out tools quickly and rely on employees to figure out best practices. Readers’ accounts suggest that the latter approach increases risk and frustration. Where governance is weak, workers become de facto compliance officers, responsible for catching errors without having the authority or resources to fix underlying issues. Where governance is strong, employees can focus on judgement rather than constant triage.
The letters also reveal a shift in accountability. AI can generate content, but someone must own the decision. Readers describe how responsibilities are being redistributed: managers ask for different kinds of evidence, teams maintain logs of prompts and sources, and review processes become more formal. In some workplaces, the human role evolves into a “quality gatekeeper” who ensures that outputs meet legal, ethical, and brand standards. In others, the human role becomes a “workflow designer,” tasked with shaping how AI is used so that it produces reliable results within defined boundaries.
This is where the postbag becomes especially instructive: concerns about accuracy and accountability are not theoretical. Multiple correspondents raise practical questions about reliability—what happens when AI outputs are wrong, incomplete, or biased; how to handle edge cases; and how to prevent errors from spreading through downstream systems. Readers describe scenarios where AI summaries omit critical details, where generated text inadvertently contradicts policy, or where automated classification misroutes sensitive requests. These incidents may be relatively rare, but their impact can be outsized, particularly in regulated environments.
The letters also show that change management is becoming a central workplace issue. Adoption is not simply a technology rollout; it is a negotiation over how work will be done. Readers describe resistance that is sometimes emotional—fear of being replaced, fear of being blamed, fear of losing professional identity. But they also describe resistance that is rational: if AI is introduced without clarity about standards, employees cannot reliably do their jobs. Several correspondents argue that organisations need to communicate not only what AI will do, but what it will not do, and how performance will be measured. Without that, workers experience AI as an unpredictable supervisor rather than a tool.
A unique take emerging from the postbag is the idea that AI is changing the “shape” of expertise. In many professions, expertise has traditionally been demonstrated through the ability to produce outputs from scratch: writing reports, building analyses, drafting proposals, or responding to complex queries. With AI assistance, the ability to produce a first draft becomes less distinctive. What becomes more valuable is the ability to define the problem, choose the right approach, and ensure that the output aligns with real-world constraints. Readers describe a growing premium on contextual judgement—knowing what matters, what is missing, and what should be questioned.
This shift can be empowering for experienced workers. Some correspondents say AI frees them to focus on higher-value tasks: strategy, stakeholder communication, and problem-solving. Others say it threatens to deskill parts of their role by automating the early stages of work. The tension is not resolved by the presence of AI alone; it depends on how organisations redesign roles and incentives. If AI is used to expand capacity without expanding support, workers may feel squeezed. If AI is used to elevate judgement and reduce drudgery, workers may feel more effective.
The postbag also suggests that AI adoption is creating new forms of collaboration. In some teams, employees work alongside AI systems as if they were junior colleagues: they provide instructions, review outputs, and refine. In other teams, AI becomes a shared resource that multiple functions rely on, which raises questions about consistency and governance. Readers describe the need for common standards—shared templates, shared definitions, and shared evaluation methods—so that AI outputs do not vary wildly depending on who prompts the system. This is particularly important in customer-facing contexts, where inconsistency can damage trust.
Another insight from the letters is that AI’s benefits are often tied to data readiness. Where organisations have clean internal documentation, well-structured knowledge bases, and clear processes, AI performs better and adoption feels smoother. Where data is fragmented, outdated, or poorly governed, AI outputs become harder to trust. Readers describe the unglamorous work behind successful deployments: curating knowledge sources, updating policies, and building feedback loops so that the system improves over time. In that sense, AI is not only a technical upgrade; it is a forcing function for operational hygiene.
The postbag also touches on the human side of adaptation: how workers learn. Many correspondents describe informal learning—trial and error, sharing prompt techniques, comparing outputs, and building personal checklists. But they also describe formal learning initiatives, including workshops and internal communities of practice. A recurring message is that learning is faster when employees have access to safe environments where they can test AI without risking real customer harm or compliance breaches. When organisations provide sandboxed tools and clear guidance, adoption becomes a skill-building process rather than a gamble.
Concerns remain, and readers do not shy away from them. Accuracy is one. Another is accountability: if AI generates an error, who is responsible—the employee who used the tool, the manager who approved its use, or the vendor who