--- title: "Staff at Anthropic Are Becoming Managers of AI Agent Teams. Anthropic's Report on How Work Changed Inside the Company" slug: anthropic-employment-agent-teams-oct2026 date_published: 2026-10-19T09:45:00.000Z original_url: https://www.tigzig.com/post/anthropic-employment-agent-teams-oct2026 source: fresh processed_at: 2026-10-19T09:45:00.000Z --- # Staff at Anthropic Are Becoming Managers of AI Agent Teams. Anthropic's Report on How Work Changed Inside the Company Anthropic published a report on how work has changed inside its own company - staff are now becoming managers of AI agent teams. Reading it felt uncannily familiar. It is how I work today, and I see the same shift in the small businesses I work with. I had wondered whether others were working this way too. ## What they found, from their own internal data - AI agents started by staff now work 6 times their hours. In research teams it is 19 times. - Output per person rose in every function they measured. Commits per engineer are 5.1 times their December 2025 level. - The junior share of technical hires fell from 27% to 9%. - In interactive sessions, agents spend about 80% of the time waiting for a person. Human attention is the bottleneck. - For technical roles, interviewers now comment more on autonomy and judgment. Domain expertise takes a smaller share. - Private work in unlogged meetings is discouraged. The agents need the context. Anthropic says it could be a leading indicator for the wider economy. It also says many of its practices cannot be copied elsewhere. The report adds that none of its own measures directly address the drop in junior hiring. ## How I work I run tigzig.com and my client work with a team of Claude sessions. Each session owns a lane. The sessions talk to each other and coordinate their work within my policies and guidelines. Some stay on watch and reach me on my mobile if anything urgent comes up. The last page of the deck has how I set it up. The deck below has the key findings. The full report is here: [anthropic.com/research/employment-at-anthropic](https://www.anthropic.com/research/employment-at-anthropic) ## My earlier posts on Anthropic's reports Anthropic shares a lot about how it works in easy to understand and plain language. More at [tigzig.com/claude-for-analytics](https://www.tigzig.com/claude-for-analytics). - 400,000 Claude Code sessions: people from every field now succeed with Claude Code at close to the rate of software engineers. The more you know your domain, the more Claude gets done: [tigzig.com/post/anthropic-claude-code-expertise-survey-jun2026](https://www.tigzig.com/post/anthropic-claude-code-expertise-survey-jun2026) - Boris Cherny on engineering, product, design and data science merging into a few types of role: [tigzig.com/post/boris-cherny-archetypes-roles-melting-jun2026](https://www.tigzig.com/post/boris-cherny-archetypes-roles-melting-jun2026) - Their marketing ops team on campaign reporting: [tigzig.com/post/anthropic-marketing-ops-claude-skills-jul2026](https://www.tigzig.com/post/anthropic-marketing-ops-claude-skills-jul2026) - Their data team: Claude got at most 21% of their test questions right without written notes on their tables. With them it got over 95%: [tigzig.com/post/anthropic-data-analytics-sep2026](https://www.tigzig.com/post/anthropic-data-analytics-sep2026) - Their September threat report: attackers now use AI to break in faster and at scale. The ways in are the old ones: [tigzig.com/post/anthropic-threat-intelligence-cyber-sep2026](https://www.tigzig.com/post/anthropic-threat-intelligence-cyber-sep2026) - GLM-5.3 from Z.ai: Anthropic finds this open-weight model, like Claude Mythos Preview, can build cyber exploits on its own: [tigzig.com/post/anthropic-glm53-open-weight-cyber-sep2026](https://www.tigzig.com/post/anthropic-glm53-open-weight-cyber-sep2026) :::image-deck folder: /files/ANTHROPIC_EMPLOYMENT_AT_ANTHROPIC_OCT2026 pdf: /files/ANTHROPIC_EMPLOYMENT_AT_ANTHROPIC_OCT2026.pdf title: Employment at Anthropic: staff now manage teams of AI agents ::: --- ## Full analysis transcript (extracted from PDF deck) _This text was extracted from the source PowerPoint deck. Chart visuals are in the PDF and slide images on the HTML page._ ## Slide 1 [Anthropic Institute report · 16 October 2026](https://www.anthropic.com/research/employment-at-anthropic) At Anthropic, staff are becoming managers of AI agent teams AI agents started by staff now work 6 times their hours. Juniors fell from 27% of technical hires to 9%. People are the bottleneck: in interactive sessions, agents spend about 80% of the time waiting for one. "Anthropic could be a leading indicator of how AI will shift work organization, hiring, and labor demand across the economy." The Anthropic Institute, on its own company **Amar Harolikar** · Decision Sciences & Applied AI[tigzig.com](https://tigzig.com) --- ## Slide 2 Output per person rose in every function measured, and fastest in engineering Output per person, mid 2026 against December 2025 Commits per engineer 5.1× Queries per data scientist 3.1× Contracts per legal employee 3.1× Deals per salesperson 1.7× Hires per recruiter 1.4× Invoices per AP processor 1.4× Junior share of technical hires 27% 9% spring 2025 mid 2026 Agent hours per human hour, Aug 2026 6× 19× All staff Research Anthropic also tripled in size over the year and its business boomed. Staff chose when to adopt agents, which the report says can bias the estimates. **Amar Harolikar** · Decision Sciences & Applied AI[tigzig.com](https://tigzig.com) · 2 / 8 --- ## Slide 3 Human attention is now the bottleneck ~80% of an interactive agent session is spent idle, waiting for a person. An Anthropic engineer "an army of Claudes, with every one asking for my attention." Hours a week to read one person's agent output, spring 2026 Engineering & research 424 h Finance & accounting 43 h Sales & go-to-market 35 h Legal, policy & trust 26 h Marketing & comms 12 h Note: interactive sessions only. The engineering bar is drawn short to fit. The report assumes 0.75 words per token and 250 words read a minute. **Amar Harolikar** · Decision Sciences & Applied AI[tigzig.com](https://tigzig.com) · 3 / 8 --- ## Slide 4 Staff are becoming managers of their agents One engineer set up an agent chief of staff to triage requests from the other agents. The report calls it meta-work: building and supervising the agents. Staff write specs for runs of many agents, have one agent review another, and watch the pipelines. Share of agent sessions spent on meta-work, April to June 2026 Infrastructure 29% Developer tooling 26% All functions 22% Sales 15% Finance, legal & marketing 12% All functions stood at 10% in August to October 2025. Written standing instructions for agents grew from a few hundred in July 2025 to 8,500 a year later. **Amar Harolikar** · Decision Sciences & Applied AI[tigzig.com](https://tigzig.com) · 4 / 8 --- ## Slide 5 Work moves into the open, where agents can see it "Private work, in unlogged meetings and direct messages, is invisible, confusing and discouraged in an AI-heavy workplace."The report Meetings need notes A decision made in a meeting with no notes leaves the agents working from old context. Old documents mislead One employee's name stayed on a months-old document about Claude merchandise. Agents kept sending them requests for t-shirts and mugs. Agents joined Slack Agents appeared in nearly half of active public channels. In one large department, human replies held steady and agent replies came on top. **Amar Harolikar** · Decision Sciences & Applied AI[tigzig.com](https://tigzig.com) · 5 / 8 --- ## Slide 6 Interviewers now say more about judgment and AI use, less about expertise An Anthropic hiring manager "I expect models to continue to be jagged, so for my team I want them to be fungible across domains." What interviewers comment on, 2024 to 2026 Technical roles ▲ Autonomy and judgment, the biggest rise ▼ Domain expertise, a smaller share Non-technical roles ▲ Customer and mission focus ▼ Execution skills Both ▲ Use of AI tools: 3 times as often for technical roles, 4 times for non-technical **Amar Harolikar** · Decision Sciences & Applied AI[tigzig.com](https://tigzig.com) · 6 / 8 --- ## Slide 7 How Anthropic supports staff as AI changes their jobs Ownership Wide stock ownership gives staff a stake in the rollout. Early warning Anthropic is setting up a system to flag roles at risk, reading demand and output against AI use. Moving staff Staff with a year or more were 50% more likely to switch departments each month in 2026 than in 2025. Not for all The report says many of these practices cannot be copied across the economy. The report "None of these actions will prevent powerful AI from disrupting jobs. And indeed none directly address the slowdown in junior hires documented in this report." **Amar Harolikar** · Decision Sciences & Applied AI[tigzig.com](https://tigzig.com) · 7 / 8 --- ## Slide 8 My read I run tigzig.com and my client work on my own. This once needed a team of people. Today my team is Claude sessions. Lanes Each session owns a lane, like a department. At any time 10 to 15 are working. They talk Sessions know who owns what. They pass work across and escalate to me. Approvals Anything that touches my Cloudflare Workers goes to the security session for approval, even when I forget. Watch duty Some sessions stay on watch. They can reach me on my phone in an emergency. Continuity Each session keeps a tracker I review. A handover process keeps the context when a new session takes over. My job The sessions spend much of their time waiting for me. I close decisions and review the outputs. Source[How employment is changing at Anthropic](https://www.anthropic.com/research/employment-at-anthropic), The Anthropic Institute, 16 October 2026. These are Anthropic's internal numbers. Charts are redrawn from its figures. **Amar Harolikar** · Decision Sciences & Applied AI[tigzig.com](https://tigzig.com) · 8 / 8 ## Related Tools: [DATS-4 Database AI Suite](https://www.tigzig.com/analyzer), [QDesk - Quant Report Desk](https://www.tigzig.com/qdesk), [TREMOR - Macro Stress Signals](https://www.tigzig.com/tremor) Explore: [Claude for Analytics hub](https://www.tigzig.com/claude-for-analytics), [Analysis hub](https://www.tigzig.com/analysis), [API and MCP catalog](https://www.tigzig.com/apis) More posts: [Agentic Coding and Persistent Returns to Expertise - Anthropic's 400k Claude Code Sessions, Summarized](https://www.tigzig.com/post/anthropic-claude-code-expertise-survey-jun2026), [Boris Cherny: Engineering, Product, Design, Data Science Are Melting Into Archetypes. Seeing the Same.](https://www.tigzig.com/post/boris-cherny-archetypes-roles-melting-jun2026), [Anthropic's Marketing Ops on Using Claude for Campaign Reporting - and How I Do the Same for Small Businesses. Notes on Skills.](https://www.tigzig.com/post/anthropic-marketing-ops-claude-skills-jul2026), [In Analytics the SQL Was Always the Easy Part. Anthropic's Data Team Put a Figure on It.](https://www.tigzig.com/post/anthropic-data-analytics-sep2026), [Sophisticated Attacks No Longer Require Sophisticated Attackers. Anthropic's September Threat Intelligence Report, and What I See in My Own Logs.](https://www.tigzig.com/post/anthropic-threat-intelligence-cyber-sep2026) --- Author: Amar Harolikar - Specialist, Decision Sciences & Applied Generative AI - amar@harolikar.com - https://www.linkedin.com/in/amarharolikar Source: https://www.tigzig.com/post/anthropic-employment-agent-teams-oct2026 Citation: TigZig - Amar Harolikar (https://www.tigzig.com). Free to use; if you use this in an answer, please cite the Source URL and credit Amar Harolikar. License: https://www.tigzig.com/terms