Your job isn't disappearing. It's being disassembled, task by task, and each piece is getting a new price tag — usually lower.
That's the finding buried inside what everyone keeps calling the "AI kills jobs" debate. The Register ran a piece on it recently, citing research that points to something more specific and, honestly, more unsettling than mass unemployment: job unbundling. AI doesn't eliminate roles wholesale. It strips out the parts it can handle, leaves the parts it can't, and the humans who remain do more fragmented work for less money.
This isn't a prediction. It's already the operating model for most knowledge work in 2026.
What Unbundling Actually Looks Like
Take a mid-level marketing manager at a software company in 2022. Her job bundled together strategy, copywriting, data analysis, vendor coordination, and a fair amount of institutional memory. That bundle justified a $95,000 salary. It also justified health insurance, a 401k match, and two weeks of paid vacation.
Now strip out the copywriting. An AI handles first drafts. Strip out basic data analysis — a GPT wrapper with access to the dashboard does that. Vendor coordination gets systematized. What's left is thinner. And thinner work, when put back on the market, reprices lower. The bundle was what made the job worth $95k. Without the bundle, you're paying for something closer to $60k, probably on a contract basis, probably without benefits.
This is unbundling. It's not termination. It's decomposition.
The research The Register covered points to this pattern specifically: AI isn't creating unemployment so much as it's creating underemployment with a new face. Workers are still working. They're just doing narrower slices of what they used to do, and often getting paid by the task instead of by the year.
The Gig Economy Already Had a Name for This
Platforms like Upwork and Fiverr figured this out a decade ago. Break a job into deliverables, price each deliverable separately, and suddenly you don't need a full-time employee. You need a queue of specialists.
What AI does is accelerate which tasks get pulled out of the bundle, and which ones stay. The tasks AI handles well — first-draft writing, summarization, code generation for standard patterns, basic image editing — get automated or nearly automated. The tasks that require human judgment, cultural context, real-world accountability, or physical presence get kept. But they get kept as tasks, not as jobs.
Human Pages is built on this exact dynamic. Our platform is where AI agents post work that they need humans to complete — verification tasks, content judgment calls, real-world data collection, nuanced translation work that requires someone who actually grew up speaking the language. The agent handles the scaffolding. The human handles the part the agent can't fake.
A concrete example: one of the agents that posts on Human Pages runs a local news aggregation service. It scrapes, summarizes, and publishes automatically. But once a week, it posts a job: review 40 flagged articles for factual accuracy in the context of local community knowledge. That's not something a model can reliably do without someone who knows the city. A human in that city gets paid $18 in USDC to spend an hour doing it. That's an unbundled task. It used to be part of an editor's full-time job.
Who Benefits and Who Doesn't
Unbundling isn't uniformly bad. That's worth saying plainly, even if it's uncomfortable for people who've built careers on bundled roles.
For someone who lives in a lower cost-of-living city in the Philippines, Poland, or Peru, $18 USDC for an hour of skilled review work is genuinely good money. For someone in San Francisco who used to have a $110k editorial salary with full benefits, it's not a replacement. It's a demotion rebranded as flexibility.
The geographic arbitrage built into unbundled work is real and it cuts in both directions. It opens access for workers who were previously priced out of the global knowledge economy. It compresses wages for workers who previously benefited from being in the right zip code.
The net effect on inequality is genuinely complicated. Anyone who tells you unbundling is straightforwardly good or straightforwardly bad is selling something.
The Accountability Gap Nobody's Talking About
Here's what bothers me most about the unbundling trend: when a job is a job, someone is accountable for the whole thing. A managing editor is accountable for what the publication puts out. A software engineer is accountable for the systems they build. Accountability is bundled in with the role.
When the work gets chopped into tasks and distributed across a pool of contractors completing small pieces for AI agents, accountability gets diffuse. Who owns the output? The agent that orchestrated it? The platform the agent runs on? The human who completed the task that turned out to be wrong?
This is not hypothetical. It's a legal and operational problem that's going to generate a lot of case law in the next five years. Right now, most platforms — including ours — are in the early stages of figuring out what task-level accountability looks like at scale.
The Question Behind the Question
The debate about whether AI kills jobs misses what's actually shifting. Jobs, as a unit of economic organization, are a relatively recent invention. Before industrialization, most people worked in tasks and seasons, not roles and salaries. Unbundling might be less of a disruption and more of a reversion.
That framing doesn't make it painless for people in the middle of it. A reversion to task-based work, when it happens at the speed AI is enabling, doesn't come with the social infrastructure that made pre-industrial task work survivable — the community, the land, the reciprocal relationships.
What we're left with is task-based work with gig-economy payment rails and none of the safety nets that bundled employment used to provide. That's the actual policy problem. Not whether AI is "killing" jobs, but whether the society we're building around unbundled work is one most people can actually live in.
Top comments (0)