WorkerKitNotes on AI workers
How AI workers are built, run and paid for: kits versus plugins and skills, managed runtimes, schedules and triggers, model choice and what a run costs.
Read the index as markdownAll posts
- Cost per acceptable AI result
Measure AI worker value using accepted outcomes, including execution, retries, human review, correction and maintenance in the total cost.
8 min read - From AI tools to an AI workforce
How people, AI models and workers operate together, with managed cloud infrastructure, shared knowledge and company control over each job.
9 min read - Claude Code MCP Tasks for long-running jobs
Claude Code now runs long MCP tool calls as tasks. The server returns a handle and Claude keeps working. How it works, its limits, and where workers fit.
13 min read - AI agents vs RPA: choose by process step
Compare AI agents, RPA and workflow automation by the work each step needs. Design a support process with clear handoffs, review and execution records.
8 min read - Enterprise AI automation: a practical blueprint
Build enterprise AI automation around defined jobs, app permissions and measurable results, with practical use cases and a process brief for your team.
9 min read - Evaluating enterprise AI automation platforms
Compare enterprise AI automation platforms on one real job, with mandatory controls, acceptance tests, total operating cost and a clear decision record.
9 min read - What is Claude Duet? Features, access and limits
Claude Duet is Anthropic's proactive agent app. What it does, who can get it, how usage limits work, what it asks before acting, and where AI workers fit.
10 min read - Claude Duet vs OpenAI Dots vs Meta Muse
Compare Claude Duet, OpenAI Dots and Meta Muse from their launch material: computer, channels, work while away, approvals, access, cost and connections.
9 min read - Hark Pro vs Meta Muse vs Instinct vs Grok Bot
Compare Hark Pro, Meta Muse, Instinct and Grok Bot from their own launch pages and terms: computers, memory, approvals, pricing and connections.
13 min read - OpenAI Decisions API: the public beta explained
OpenAI's Decisions API is in public beta: typed questions on gpt-6-luna at $0.10 per 1M input tokens. How it works, what it costs, and how it compares with Jev.
10 min read - OpenAI Dots vs Meta Muse vs Grok Bot
Compare OpenAI Dots, Meta Muse and Grok Bot using their launch announcements: computers, channels, connected apps, access and approvals.
7 min read - OpenAI Decisions API vs Jev, field by field
OpenAI Decisions API vs TypeSafe Jev, now that both contracts are public. Question types, confidence, refusals, images, limits, data terms and a 2.4x price gap.
11 min read - What are OpenAI Dots? Features and access
OpenAI Dots are always-on agents in ChatGPT. Learn how to get a dot, what it can do, launch pricing and limits, and how its approvals work.
7 min read - Decision model vs agent: the two halves
Almost every job handed to an AI agent is two jobs stapled together: a judging half and a working half. They want different machinery.
13 min read - Jev: TypeSafe's System One model explained
Jev is TypeSafe's System One model. It answers typed questions with probabilities and writes no text at all. What it does, costs, and cannot do.
13 min read - Screen cheap, spend expensive: agent triage
A worker that reads 400 items to act on 9 bills you for reading 400. Three ways to put a cheap screening pass in front of the expensive work.
11 min read - Connecting Meta's Muse Code to your AI workers
Meta's Muse Code shipped with MCP support. Point it at the WorkerKit MCP server and the coding agent can read, run and schedule your AI workers.
9 min read - Muse Code connectors: adding MCP servers
How connectors work in Meta's Muse Code: the mcp_servers settings format, both transports, the sandbox caveat, and a first server worth adding.
9 min read - A managed runtime for unattended AI jobs
An unattended AI worker is not a model deployment, it is a job deployment. What a managed runtime standardizes is everything that job needs to keep running.
11 min read - Bot directory: what to check before installing
A bot directory is a browsable catalog of bots you can install. Here is what a listing has to tell you, and the five checks worth making before you run one.
8 min read - Orchestrating an SEO agentic workflow
One worker finds what deserves a page, one drafts it, one publishes it. How to chain three AI workers into an SEO pipeline, and where the human gate belongs.
12 min read - AI agent vs AI worker: where the line is
An AI worker is an agent with a persistent job, permissions, limits and run record. Compare the terms without confusing scheduled work with ownership.
8 min read - Case study: a worker that texts real people
Appointment Reminder Texter is the shape where a mistake reaches a stranger. What contact rules actually do, and why this one earns a narrow send stage.
6 min read - Case study: a morning brief that reads only
A walkthrough of Executive Daily Briefing, the read-only shape: three apps, no write access anywhere, and why that makes it the right first worker.
6 min read - What an unattended AI fleet needs to fail well
Unattended agents fail plausibly rather than loudly. Five design choices decide whether a bad run costs an hour or a quarter, and none is model quality.
6 min read - How to pick a model for an AI worker
Benchmarks measure somebody else's task. The only comparison that counts is the same run, on your instruction and your data, judged by you.
6 min read - How to publish a kit others can run
A worker that works for you is not yet a kit. What has to be true before publishing, and what the permanent address commits you to afterwards.
6 min read - Why an instruction is not a safety control
Telling a model what not to do is a request it weighs against everything else. At unattended scale, a rule that holds 99.5% of the time fails every month.
7 min read - How to roll out an AI worker safely
Four stages from read-only to autonomous, what to check before each promotion, and the signals that mean go back a step rather than push on.
7 min read - Schedule or webhook? How an AI agent starts
Polling burns capacity discovering nothing happened. Triggers fail silently. The right answer for most unattended workers is both, and the split is not obvious.
7 min read - What to automate first, and what not to
Most teams pick their worst job to automate first. Three tests that predict whether a job will work unattended, and the ones that reliably fail.
7 min read - When a kit is 80% of what you need
The kit does most of the job and gets one thing wrong. What you can change after installing, what you cannot, and when to start from scratch instead.
7 min read - When a worker writes to your CRM
A wrong CRM field looks recoverable and is not. What changes when an AI worker writes into a system other people read and act on later.
7 min read - How to write an instruction that holds
A prompt that works in testing drifts by run four hundred. What separates a chat prompt from a standing instruction an unattended worker can run on.
7 min read - Inbound lead qualification with AI workers
How Inbound Lead Qualifier reads an inquiry, updates the CRM and drafts a reply, with event triggers, review boundaries and cost planning.
9 min read - Case study: a support inbox triage worker
A walkthrough of deploying the Support Inbox Triager kit: the access it needs, the schedule, the safety rules, what it costs and where it should stop.
8 min read - Agent skills vs plugins vs MCP vs kits
Skills teach a technique, plugins package it for a client, MCP connects your systems, and kits are workers that run without you. Where each boundary sits.
10 min read - What an AI worker actually costs to run
AI worker pricing runs on three meters: the plan, the tool calls and the model tokens. What each measures on WorkerKit, with real figures and worked examples.
10 min read
About this blog
Writing on how to turn a defined job into an AI worker your company keeps: its method, its app permissions, its spending limits and its run record. Start with the path from AI tools to an AI workforce, then use the job guides and evaluation articles to put that approach to work.
What gets written about here
The category. Where kits sit next to skills, plugins and MCP, and what a persistent worker adds around a defined job.
Building a workforce. First worker, connected workers, company workforce: who owns the method, how workers hand off work, and when adding another worker is justified. The inbound-sales guide turns that sequence into a procedure using real kits.
Enterprise AI automation. Start with a process blueprint, then use the platform evaluation worksheet to compare authority, outcomes and operating cost. The AI agents versus RPA comparison helps assign each step; the support-triage guide puts those decisions into an inbox-to-ticket procedure.
Running workers. Managed runtimes, schedules versus webhook triggers, and what an unattended fleet needs in order to fail safely.
What it costs. The three separate meters: the plan, tool calls, and model tokens at provider list price with no markup, paid from a wallet or your own key. Cost per acceptable result adds evaluation, failed attempts, human review, correction and maintenance to the buying decision.
Judging before acting. What a System One model is, where the judging half of a job differs from the working half, and why a cheap screening pass in front of the expensive work changes which backlogs are worth automating at all.
Keeping it safe. Why an instruction is a request to a model and an app firewall, contact rules and redaction are not.
Getting started well. Which job to automate first and which to leave alone, how to roll a worker out in stages from read-only to autonomous, and how to write an instruction that still holds on the four hundredth run.
Building and publishing. What to do when a kit is most of what you need, and what has to be true before you publish one others will install.
Walkthroughs. Four real kits taken end to end, one per risk shape: a read-only brief, a triager that drafts, an event-driven qualifier, and one that texts real people. Each states the access it needs, what it costs and where it should stop.
FAQ
What is an AI worker?
An AI worker is an AI agent configured for a defined job with its own app permissions, limits and run record. It runs on demand, on a schedule or on a supported event. The job's configuration persists after a run ends, and each run leaves a receipt.
How do I start with an AI worker?
Browse worker kits and the job guides, then request access. WorkerKit accounts open by invitation. Once invited, choose a kit, connect the apps it needs, review its permissions and limits, and run a bounded first check. Setup effort depends on the job and the accounts involved.
Which AI models can run these workers?
Language workers can use supported models from different providers. You can change the worker's model or use a per-run override to compare results on the same job. A starting configuration is not proof of model fit: compare quality and total cost per acceptable result on your own work.
Is there a free plan?
WorkerKit has a Free shared cloud plan; account access is currently by invitation. Review the plan limits, execution seats and separate usage charges on /pricing, then join the waitlist if you need an account.
Can an AI agent read these posts without running JavaScript?
Yes. Each post page carries its full article in the initial HTML, as well as structured data. /blog.md indexes every post with its description and dates; it is an index, not a full-text copy of every article.