The future of work
Your weekly briefing on AI and human collaboration
ISSUE 016 | 19 OCTOBER 2026 | WEEK 13–19 OCTOBER 2026
This week in perspective
The shift to agentic AI is increasingly visible in how work is organised, not simply how tools are used. Anthropic reports agent activity exceeding employee hours, while OpenAI sees rapid growth in agentic usage across its EU sample. But neither signal proves lasting enterprise value. Alongside new cyber risks, persistent skill shortages and growing work intensity, this week’s evidence raises a sharper question for CHROs: who designs, supervises and benefits from work completed by people and agents together? The challenge is to measure outcomes and human capability, not adoption in isolation.
Top stories
01 AI-agent hours reshape the work at Anthropic
Anthropic’s 16 October analysis of its own workforce found that by August, interactive agent hours were six times employee working hours. The share of junior technical hires declined as agent use expanded, while employees increasingly developed monitoring tools to manage agent output.
Why it matters: Rising agent capacity is making human attention, judgement and verification central constraints on productivity. Findings from one AI company should not be assumed to describe every employer.
SOURCE [1] Anthropic · 16 October 2026 · Read source
02 EU agentic AI use surges—but outcomes remain unproven
On 14 October, OpenAI reported that Codex and ChatGPT Work generated more than half of output tokens in its sampled EU enterprise accounts. Weekly active users of those tools among EU small and medium businesses grew more than eightfold since May.
Why it matters: These are platform usage indicators, not independent evidence of productivity gains. CHROs should measure redesigned workflows, human contribution, quality and business outcomes.
SOURCE [2] OpenAI Economic Research · 14 October 2026 · Read source
03 CEOs prioritise AI-enabled deception and agent data risks
KPMG’s 15 October US CEO Outlook found that 78% of 400 surveyed US CEOs cited AI-powered social engineering and targeted attacks as drivers of increased cybersecurity investment. Another 73% identified data and privacy risks linked to AI agents.
Why it matters: AI governance must cover both external manipulation of people and internal agent permissions. Human verification and bounded authority belong in the operating model.
SOURCE [3] KPMG US · 15 October 2026 · Read source
04 OECD highlights persistent mismatches in workforce skills
The OECD published its updated Skills for Jobs analysis on 14 October, mapping shortages and surpluses across 38 countries. Its analysis highlights continued pressure in highly skilled occupations alongside contrasting conditions in other parts of labour markets.
Why it matters: External shortages should inform, not replace, internal skills intelligence. Re-engineering tasks, reskilling and redeployment may unlock capability before organisations recruit externally.
SOURCE [4] OECD Publishing · 14 October 2026 · Read source
05 AI-enabled connection coexists with growing work intensity
Jabra’s WorkLife 2026 research, publicised on 13 October, surveyed more than 6,000 knowledge workers. It reported increased collaboration for 42% of respondents and associated frequent AI use with stronger connection, even as workloads and expectations expanded.
Why it matters: AI may change collaboration without easing total workload. Leaders should track employee experience, effort and connection alongside time saved; survey associations do not establish causation.
SOURCE [5] Jabra WorkLife · 13 October 2026 · Read source
ALISON’S VIEW
The overlooked work behind the AI productivity story
Alison Ettridge | Founder, AmpliTask
I find this week’s evidence a useful challenge to a familiar assumption: more AI activity equals better performance. Anthropic’s own experience suggests that as agent output expands, humans have to create new ways to monitor, validate and coordinate it. That is work, even when it does not appear in the automation business case.
Meanwhile, persistent skills shortages and growing work intensity remind us that organisations cannot simply add agents to yesterday’s processes and expect better outcomes. We have to redesign tasks, hand-offs, accountability and learning together.
For CHROs, this means treating human judgement and capability as assets to amplify, not overhead to eliminate. The critical measure is not how much work an agent completes, but whether the whole human-plus-agent system creates more value—safely, sustainably and with people in the lead.
Themes to watch
Human attention is becoming a scarce resource: Anthropic’s expanding agent output and OpenAI’s usage growth [1–2] suggest rising coordination demands. Measure verification workload and actual outcomes before treating activity as productivity.
Trust and capability must develop together: KPMG’s cyber concerns [3], OECD shortages [4] and Jabra’s workload findings [5] point towards a need for safe authority boundaries, reskilling and sustainable work design.
Key actions for workforce leaders
Three decisions to take from this week’s evidence:
| Action and intended outcome | Owner and timing | Evidence of progress |
| Map one agent-heavy workflow: capture execution, supervision, escalation and rework tasks; identify where value is created. | CHRO + process owner Within 30 days | Signed task-and-decision map; baseline hours, quality, exceptions and human oversight effort. |
| Test AI-enabled deception and access: require out-of-band approval for sensitive requests and establish bounded agent permissions. | CISO + AI governance lead Within 30 days | Documented control test, accountable owner and remediation actions for gaps. |
| Integrate skills and workload evidence: compare critical gaps, internal redeployment and team connection before expanding AI. | Head of Workforce Planning Within 45 days | Top three gaps and redeployment options; workload/connection measures and owners. |
Sources and further reading
[1] Wilmers, Catipon, Massenkoff & McCrory · How employment is changing at Anthropic · Anthropic · 16 October 2026 · Source
[2] OpenAI Economic Research · 14 October 2026 · EU agentic AI use surges—but outcomes remain unproven · Source
[3] KPMG US · 15 October 2026 · CEOs prioritise AI-enabled deception and agent data risks · Source
[4] OECD Publishing · 14 October 2026 · OECD highlights persistent mismatches in workforce skills · Source
[5] Jabra WorkLife · 13 October 2026 · AI-enabled connection coexists with growing work intensity · Source
Continue the conversation
Are your AI programmes creating better work—or merely more automated activity? Reply to alison@amplitask.com to explore an AmpliTask work re-engineering diagnostic that connects tasks, human capability, agent oversight and value creation.
AmpliTask · Maximising human impact in the future of work.
http://www.amplitask.com | alison@amplitask.com | https://www.linkedin.com/company/amplitask/home
© 2026 AmpliTask · AmpliTask Ltd, Brynderwen, Talyllyn, Brecon, LD3 7SY, UK.