You know the feeling. You need a simple answer, but it requires data from another team. So you write a request, explain context, wait for a reply, schedule a meeting, discuss, wait again, get partial answer, clarify, wait more… By the time you get the data, the problem has changed. That’s the handoff tax. And AI is about to eliminate it.
OpenAI recently analyzed over 800,000 ChatGPT messages from U.S. workers. The finding? 43.5% of work-related queries involve tasks that belong to a different job title. Customer service reps are running marketing analysis. Designers are debugging code. HR is doing finance. The boundary between roles is dissolving — not because jobs are disappearing, but because the handoffs between them are evaporating.
The first cost AI reduces isn’t execution — it’s the cost of handing work from one person to another. This changes everything about how we design software.
The Old Way: Software for Job Titles
For decades, we built products around roles. CRM for sales. Jira for engineers. Tableau for analysts. We asked: “Who is this user?” and then gave them a fixed set of features. But AI doesn’t care about your job title. It cares about what you’re trying to accomplish right now.
A product manager might need to analyze interview transcripts in the morning, clean event data at noon, and prototype a small tool by evening. If your software only sees a “product manager,” you’ll miss most of what they actually do.
The New Unit: Task, Not Role
Every task has six components: a goal, inputs, constraints, capabilities, a deliverable, and acceptance criteria. When you organize software around those, you stop forcing users to piece together workflows from scattered modules.
Consider: “Prepare a management briefing for next week’s product launch.” This single task cuts across product, marketing, sales, data, and design. The user doesn’t want five different tools — they want one coherent path to the result.
Your product shouldn’t be a toolkit. It should be a task-completion machine.
Five Things You Must Change
1. Onboarding: Ask what they’re doing, not who they are. Stop leading with “Select your role.” Instead: “What’s the one thing you need to finish this week?” Role still matters for context and language, but it shouldn’t lock the user into a narrow set of capabilities.
2. Features: Modularize, don’t bundle. A designer who occasionally needs to test a technical prototype shouldn’t have to buy a full engineering suite. Break capabilities into composable blocks — search, analyze, generate, code, validate, publish — and let the task assemble them.
3. Permissions: Tie to tasks, not just accounts. This is the dangerous one. Just because someone can do a task doesn’t mean they should have permanent access to everything. Check: What can this task see? What can it do? How long does it last? For irreversible actions (payments, publishing, sensitive data), require explicit approval.
4. Deliverables: Make them work beyond the chat. The worst outcome of a cross-task AI is a beautiful answer that can’t be used. If a marketer asks for a data analysis, the output should be a verifiable, reusable dataset — not a paragraph of interpretation. The deliverable must be ready for the next person in the chain.
5. Metrics: Measure handoffs eliminated, not just messages sent. One user who completes a task that previously required three departments is worth more than a hundred users chatting idly. Track task completion rate, deliverable adoption, and the time from problem to usable result.
Four Questions Every Product Manager Should Ask
- How many people does your user need to talk to just to get their core job done?
- Which of those handoffs exist only because of tool or skill gaps — not because of real responsibility barriers?
- Where can AI let the initiator go further before they need a specialist?
- Is your product just adding a search bar to old software, or are you rebuilding the task structure?
If you can’t answer the last question, you’re building a chatbot, not a product.
What This Means for Small Teams vs. Big Companies
OpenAI’s data shows that cross-task AI usage is slightly higher in 2–5 person workspaces (18.9%) than in 100+ person companies (16.3%). Small teams have always had to wear multiple hats. AI amplifies that. The opportunity there is to help them fill missing functions without hiring.
For large enterprises, the challenge is different: enable task crossover without breaking data governance, audit trails, and accountability. The same AI capability needs different packaging depending on organizational size.
Don’t Panic: Professionals Still Matter
This isn’t a call for everyone to become a unicorn. AI lowers the barrier to try and execute, but it doesn’t replace experience, judgment, or responsibility. A marketer can generate a SQL query, but they still need to know if the metric definition is sound. A product manager can prototype a UI, but a proper engineer must still review security and performance.
What changes is when the professional gets involved. Instead of starting every handoff from scratch, the initiator can do the heavy lifting — gather data, create drafts, build prototypes — and then pass a nearly finished product to the expert for final review and high-stakes decisions.
The low-value back-and-forth disappears. The high-value judgment gets concentrated.
The Bottom Line
OpenAI’s 43.5% figure isn’t the whole story — it’s a snapshot from one platform, one country, one time. But the direction is clear: people are already using AI to bypass organizational friction. The question isn’t whether this will happen. It’s whether your product will help them do it better, or get in the way.
Stop asking, “Who is this user?” Start asking, “What task is this user trying to complete today?” That’s the difference between software that just adds AI, and software that actually changes how work gets done.
FAQ
Q: Isn't this just a fancy way to say AI will replace jobs?
A: No. The data shows people are still doing their jobs — they're just also doing parts of adjacent jobs. Roles remain, but the boundaries between them become porous. The real risk isn't job loss; it's being stuck in an organization that still requires five handoffs to get a simple answer.
Q: Should I rebuild my entire product around tasks immediately?
A: Not overnight. Start by identifying the top three tasks your users complain about most. Map the handoffs involved. Then test a task-based flow for just one of those — likely one that crosses multiple roles. Measure the reduction in time and back-and-forth before scaling.
Q: Doesn't this mean we're all going to have to become generalists?
A: No. It means specialists will spend less time on low-value coordination and more time on high-value judgment. The role of the expert shifts from 'the person who does the work' to 'the person who validates and approves the work.' The specialist's craft becomes more important, not less — but they'll be called in at the right moment, not at the start of every request.