Install computer hardware.

Most of the output can be produced by software today.

Work that already happens inside a computer is the easiest to hand over, because the agent and the worker are operating in the same place. The constraint is permissions rather than capability: the agent needs its own scoped access, not a borrowed password.

Key facts about this work activity
How automatableMostly automatable
Kind of workWork done in software
Jobs that do it9 occupations
Tool categories that applyTools and auth, Builders, Generalist agents, Runtime and sandboxes
O*NET activityWorking with Computers (4.A.3.b.1)

What could do this work.

These are the categories of tool that address work done in software, with a few live examples from our landscape. We name the category rather than promise that a specific product does your specific task, because that promise would not be true.

Tools and auth

Tool layer giving agents over 1,000 toolkits with managed connected accounts.
Developers
Free tier
MCP runtime built around per-user OAuth delegation and audited authorisation.
Developers
Free tier
10,000-plus prebuilt actions across 200-plus connectors with managed auth.
Product teams
Custom
Developer infrastructure for custom API integrations, around 900 of them, self-hostable.
Developers
Free tier
Embedded integrations with ActionKit exposing 1,000-plus tools in one call.
SaaS vendors
Custom
Roughly 2,700 APIs behind a low-code workflow layer. Acquired by Workday.
Developers
Free tier

Builders

n8n
Visual workflow automation with roughly 70 AI nodes. Self-hostable and free that way.
Technical ops
Free self-host
The default automation layer, now with Agents, Tables and Canvas bolted on.
Everyone
From $29.99/mo
Make
Visual scenario builder over 3,000-plus apps, materially cheaper than Zapier at volume.
Ops teams
From $9/mo
Build "AI employees" in plain English. Claims 5,000-plus integrations.
Non-technical
From $19.99/mo
Teams of specialised agents sold as an AI workforce, strongest in sales.
GTM teams
From $29/mo
Visual canvas for multi-agent orchestration, aimed at technical builders.
Technical ops
From $37/mo

Generalist agents

Browses, runs code and completes multi-step tasks inside the chat everyone already has.
Everyone
From $20/mo
Long-context assistant with computer use, plus Claude Code for engineering work.
Everyone
From $20/mo
Google’s assistant, with Project Mariner for browser control and deep Workspace reach.
Everyone
From $20/mo
Autonomous agent that plans and executes long tasks in its own virtual machine.
Prosumers
From $20/mo
Super-agent bundling research, calls, slides and video generation on a credit meter.
Prosumers
From $24.99/mo
Answer engine with agentic research and shopping actions.
Everyone
From $20/mo

Runtime and sandboxes

E2B
Secure cloud sandboxes for running agent-generated code.
Developers
Free tier
Fast-provisioning sandboxes aimed at agent workloads.
Developers
Free tier
Serverless compute billed by the second, widely used for agent tool execution.
Developers
Usage-based
Model inference infrastructure with dedicated GPU capacity.
ML teams
Usage-based

Categories are matched from the kind of work, not from vendor marketing. Before committing to any of them, check that the tool can reach the system your record actually lives in. That connection, not the model, is where most of these projects stall.

Work that goes with it.

O*NET groups these under “Set up computer systems, networks, or other information systems”. In practice they tend to be done by the same person, in the same sitting.

Related work activities
ActivityJobs
Install computer software14
Configure computer networks5
Install programs onto computer or computer-controlled equipment4

Method. The activity, its taxonomy placement and the occupations that perform it come straight from the public-domain O*NET 30.3 database. The automatability band is ours: we map each of O*NET’s 41 generalized work activities to how much of its output current software can produce, assuming a person still reviews and owns the result. It is a coarse three-way judgment on purpose. A precise-looking percentage here would be invented.