concept · concepts/Agents as Employees vs Human-Agent Collaboration.md

Agents as Employees vs Human-Agent Collaboration

status: active · created 2026-06-02 · 1 source · #agents #human-in-the-loop #workflow #taxonomy

Dan Shipper's taxonomy of the two ways teams actually work with agents. The discourse predicted the first; the second is "stranger and, in my experience, more important."

Mode 1 — Agents as employees (delegated, out-of-loop)

You give the agent a job and it goes off to produce an answer, action, report, draft, or triage decision without you in the loop. Two sub-types:

Coworker agents — live in Slack, have names and jobs, get tagged to do work. Example: Every's editorial agent "Andy" collects story "nuggets" from internal Slack and turns them into digests / first-pass takes for writers.
Embedded agents — live inside a product's workflow; less flexible but powerful for repetitive tasks. Example: "Fin" in Every's customer-service platform (formerly Intercom) handling support over chat/email.

These take over "the stable, repeatable, well-framed layer of work."

Mode 2 — Human-agent collaboration (shared, semi-synchronous)

Tools like Codex, Claude Code, and "Claude Cowork" where you and multiple agents use the same computer at the same time to do complex original work an async agent can't. Shipper: these are "becoming operating systems for the work itself." You can see each task the agent is doing/thinking, interrupt anytime, and you own the start and end of every task. Its signature shape is The Human Sandwich.

Both modes "require you or another human in order to work well."

Connections

The Human Sandwich — the operating pattern of Mode 2.
Personal vs Team Agents — an orthogonal axis: who owns the agent (one person vs. a team), regardless of mode.
Agent Autonomy Spectrum — Mode 1 sits toward higher autonomy; Mode 2 is explicitly semi-synchronous.
The Infinite Backlog — Mode 1 chews repeatable work; Mode 2 handles the high-judgment remainder.
Human-in-the-Loop by Exception — cross-domain: the same automate-the-bulk / human-for-exceptions split, framed as data-pipeline design rather than agent collaboration.

Sources

The AI Daily Brief — Why Agents Still Need Humans — via Dan Shipper — After Automation.

Linked from 11

Claude Code

Anthropic's agentic coding/work harness, cited alongside Codex as a shared workspace for human-agent collaboration.

Codex

Cited (with Claude Code and "Claude Cowork") as an "operating system for the work itself" — a shared workspace for human-agent collaboration.

Dan Shipper

Two modes of working with agents (Agents as Employees vs Human-Agent Collaboration); the human-agent collaboration mode is "stranger and more important."

Dan Shipper — After Automationto read

NLW read long excerpts but explicitly skipped sections — notably Shipper's full treatment of the AGI objection ("doesn't better AI just do the expert work too?"), described as "an interesting meditation on the nature of benchmarks." Getting the primary sour…

Every

"Andy" — editorial coworker agent; collects story "nuggets" from Slack into digests/first-pass takes (Agents as Employees vs Human-Agent Collaboration).

Human-in-the-Loop by Exception

Agents as Employees vs Human-Agent Collaboration — "by exception" is a stance on how much to delegate vs. supervise.

Overview

AI and work (the founding domain) — The Bottleneck Shift (deciding what to do is the new scarce input), AI Commoditizes Expertise → Slop → demand for difference, The Human Sandwich and the two modes of working with agents, Personal vs Team Agents.

Personal vs Team Agents

Agents as Employees vs Human-Agent Collaboration — orthogonal axis: ownership (personal vs. team) vs. mode (employee vs. collaborator).

Placement and Judgment Over Raw Execution

Both modes "require a human to work well," and The Bottleneck Shift explains why: when execution is cheap, deciding what to do is the bottleneck.

The AI Daily Brief — Why Agents Still Need Humans

NLW frames the resulting work pattern as two modes (Agents as Employees vs Human-Agent Collaboration): agents as employees (async, out-of-loop — coworker agents like Every's editorial agent "Andy," embedded agents like "Fin" in support) and human-agent coll…

The Human Sandwich

This is why, per Shipper, "both of these modes require you or another human in order to work well." The sandwich is the concrete shape of the collaboration mode (as opposed to fully delegated "employee" agents).

Links to 11

In the log 1

2026-06-26 ingest

Working Heuristics — Data, Matching, Product, Org, Tooling (~30 distilled heuristics, human-supplied inline → persisted to raw/ as the human's own content, NOT agent-fetched) — 1 source, 1 entity (Medallion Architecture), 14 concepts opening the new engineering-craft domain (data modeling/pipelines, matching/identity, product/platform, org/collaboration, tickets/tooling); back-linked into 4 existing pages (The Human Sandwich, Agents as Employees vs Human-Agent Collaboration, Personal vs Team Agents, Control-Flow Integrity); updated Overview (filled the empty MOC, added domain map). PROMOTE evaluated on seed Match Process Weight to Stage and Stakes → NOT promoted (ticket/tooling-hygiene craft is adjacent but doesn't assert the weight ∝ stage×stakes thesis); cross-linked instead. Cross-domain rhymes flagged: human-in-the-loop ⇄ The Human Sandwich; allowlist ⇄ Control-Flow Integrity (default-deny); build-with-not-against ⇄ Personal vs Team Agents.