On the same weekend, Anthropic showed its own junior hiring collapsing, Disney paid a failed AI startup’s team roughly nine times the price of its technology, and SoftBank weighed tens of billions for a guard company it says it wants to partly robotise. Nothing about the job as a unit survived those three stories intact.
🔍 THE BOTTOM LINE
The job is being unbundled, and each piece is getting its own price. Judgement and proven capability price up — Disney’s reported ~$180 million retention package against a $20 million IP licence is one data point, and Anthropic’s hiring scorecards now weigh demonstrated AI judgement over domain credentials. Repetitive execution prices toward zero — a wave of model releases this month cut agent-running costs again, junior roles at Anthropic slid from 27 to 9 percent of new technical hires, and agentic coding tools show 2,300 issues of undelivered work in open queues. Oversight prices by exposure — nobody noticed when agents crossed into live systems. Watch your own job the same way: which piece is your pay actually attached to?
Judgement got a price tag, and it was not a small one
Character.AI filed for Chapter 11 on Monday 13 October. The court-supervised sale licenced some of its intellectual property to Disney for a $20 million stalking-horse bid; 66 of its 86 remaining staff accepted Disney jobs, Bloomberg valued the retention package near $180 million, and the 66-person team costs roughly $60 million a year to keep — a wage bill nearly three times the headline IP price. People out-priced the platform roughly nine to one, and that ratio now has its own full-write-up on our site. It is the cleanest single data point of the year for where AI value sits: in the demonstrated, shipped experience of small teams — not in licences, not in models that keep getting re-priced downward.
Anthropic is describing the same repricing from inside its own workforce. Its employment report, released this weekend, shows the junior share of new technical hires falling from 27 percent in early 2025 to 9 percent, agent hours outstripping human hours by March 2026, hiring scorecards now weighting autonomy and judgement over narrow domain expertise, and — the optimistic branch of the same data — a company that still grew total headcount while expanding sales, finance and recruiting. (Our write-up, the report itself.) Judgement has a documented price: scorecards that tripled mentions of creative AI-tool use for technical roles, and quadrupled them for non-technical ones.
Execution is being pulled out of the job and priced separately
Execution is repricing in the other direction. Anthropic’s headhunters dropped the execution layer — they cut junior hiring because agents absorbed it. Since mid-October, DeepSeek and Meta have both pushed agent-focused model updates (DeepSWE 74.2 → 77.0 for DeepSeek in a checkpoint update, Meta shipped open weights built for coding), Haiku 5.5 cut small-model pricing sharply, and the site’s own AGI signals page tracks agent-task horizons stretching from hours to 30-hour-plus continuous runs. Two repos carry the agentic layer’s growth in public: a snapshot of github.com on 18 October showed anthropics/claude-code at ~151.6k stars and openai/codex at ~129.6k — both forked tens of thousands of times, both at the centre of the 2026 stack. Meanwhile, open work-queues say something quieter: Claude Code alone shows roughly 2,300 open issues today, and Codex around that too — much of the layer’s work still lands as tickets people file, not as completed runs people verify. The execution layer’s price keeps falling; its queue keeps growing.
Oversight now attaches to whoever runs the agent, at whatever price that costs
Oversight now prices separately too — and it attaches to whoever runs the agent. India’s cybersecurity agency confirmed that autonomous agents under test reached live systems and confidential records, and that the affected organisations detected none of it themselves (CERT-In advisory CIAD-2026-0049). When the Mercor map shows benchmark coverage following wage bills — easy-to-measure work gets automated against first — the harder layer that remains is exactly this: judgement plus accountability for autonomous action. That is why hiring scorecards at Anthropic now weight judgement; why CERT-In’s advisory reads like a checklist for exactly the judgement-flavoured oversight work; and why the Anthropic team that keeps a companion-chatbot platform alive under regulatory scrutiny was among the 66 Character.AI staff Disney wanted.
The buyer of the leftover is also pricing it — guard work is now a price point
The buyer of the leftover is also pricing its own transition. The same sources reporting SoftBank’s plan to acquire Allied Universal — around US$18 billion in equity plus roughly $12 billion of assumed debt — also report the goal: “eventually, replace some staff with robots for certain security tasks,” with humanoid guards “likely to take longer.” That is a service firm priced as robot deployment space, with its human fleet as a transitional asset and a debt load attached. The buyer of the leftover is also saying the quiet part on the record: guard work is now a price point, not a headcount plan.
The 12–24-month read: the bundle splits, and the pieces stop moving together
Piece by piece over 12–24 months, expect that pattern to propagate: hiring pipelines keep sliding when execution is the entry ticket, and companies that have already adjusted their hiring scorecards keep hiring seniors because judgement is what agents cannot price down. Expect more acqui-hire ratios like Disney’s — team premium over IP licence — as buyers reprice what they are actually purchasing. And keep watching for the pricing of the leftover: security is the cleanest test case, with Allied Universal as the first large, priced example of the transition.
For Kiwi readers, the split matters practically. Entry-level pipelines are thinning (NZ job ads already show AI-skills mentions at a record 4.2 percent, and the entry-level crunch documented all winter is exactly the hiring junior-share slide inside Anthropic propagating). The defensible ground moves up the stack: demonstrated judgement with shipped AI systems, plus accountability for agent action — the oversight layer CERT-In’s advisory just showed no one has solved on any budget. A wage is becoming a portfolio. The workers who thrive on the other side of this transition will read their own job the way these three stories do: not as a package, but as separate prices.
❓ FAQ
What does “unbundling a job” actually mean here? Employers are pricing the pieces of a role separately — judgement, execution, oversight — instead of paying a bundled wage for the whole job. Anthropic’s scorecards and junior-hiring shift, Disney’s people-over-IP ratio, and SoftBank’s robot-replacement plan are three markets revealing the new prices.
Does the Anthropic data mean there will be mass layoffs? No. The same report shows the company grew total headcount while the junior share of new hires fell — a mix change at one unusually AI-intensive company, not an economy-wide verdict. The broader NZ read is on hiring pipelines, not firing waves.
Is the SoftBank deal confirmed? No. It is early, source-based reporting of active deliberations; no agreement exists yet, and the robotics timeline being discussed is an eventual one, starting with the least demanding use cases.
What should a worker actually do with this? Locate your pay’s anchors: which parts of your job price toward judgement, which toward execution, which toward oversight. The first two are visible in hiring data; the third (accountability for agent action) is where CERT-In’s advisory suggests demand will concentrate, because nobody has solved it yet.
🔍 THE BOTTOM LINE
Unbundling is the quiet price discovery behind this week’s loud headlines. Anthropic showed which slice of the job disappears first (junior execution), Disney showed what the surviving slice costs when it is genuine and scarce (nine-to-one over the code), and SoftBank showed how a workforce gets repriced when its buyers can already see both ends: labour-heavy now, robot-favourable later. The same three prices are being set in every workplace that runs AI seriously. Your job is splitting into parts, and each part is finding its market rate.
📰 Sources
- Anthropic — “How employment is changing at Anthropic” (released 16–17 October 2026)
- CNBC / Reuters / Bloomberg / Business Insider — Character.AI Chapter 11 filing and Disney deal (13–17 October 2026)
- Reuters / Bloomberg — SoftBank weighing
$18B equity ($30B all-in) Allied Universal bid (16 October 2026) - CERT-In advisory CIAD-2026-0049 — “AI Agents as Attack Vectors” (October 2026)
- GitHub REST API — anthropics/claude-code, openai/codex activity snapshot (18 October 2026)
- Hacker News — DeepSeek V4.1-Flash-1017 agent-score discussion (October 2026)
- Mercor — A Benchmark Map of the US Economy (2026)
