AgentKit's 45 AI Agents: What the 17 Engineer + 28 Marketing Agents Do (2026)

⚠️ Which AgentKit? Telling it apart from OpenAI AgentKit and the name-twins

Before we go deep, one thing trips up a lot of people who end up on the wrong product. This post is about AgentKit for Claude Code at agentkit.best, installed through the ak CLI - a kit of skills, slash commands, and AI agents you use directly inside Claude Code, Codex, or Copilot. It is not either of the two products below, even though the names overlap:

• OpenAI AgentKit (launched 10/06/2025) is a completely different product from OpenAI - the Agent Builder, ChatKit, and Connector Registry toolset for building agents on top of OpenAI. Check openai.com directly if that is what you actually need. It does not install into Claude Code.
• Name-twins selling similar kits: agentskit.co (a different kit, advertised at 89 agents) and agentkits.net (an aggregator/kit directory). Same-sounding name, different developer, different price, different agent roster.

Why be this careful? Because searching "agentkit agents" today returns all three mixed together: the OpenAI product, the same-named kits, and agentkit.best. Buying the wrong one is a real risk - the price, agent roster, and installation are entirely different across the three. A quick tell: only AgentKit at agentkit.best uses the ak CLI and installs straight into Claude Code. If the docs you are reading talk about an "Agent Builder" or a web drag-and-drop interface, that is OpenAI, not this post.

Short version: if you use Claude Code and want the 45-agent "lineup" described here, your destination is agentkit.best - not OpenAI AgentKit, and not agentskit.co.

What do AgentKit's 45 agents include?

According to the agentkit.best home page, the platform has 108+ skills, 95+ slash commands, and 45 AI agents. Those 45 agents break down as 17 engineer + 28 marketing. The core distinction: skills and commands are "tools" Claude uses, while an agent is a named AI role you call by name (for example, "let tester write the tests," "have planner draft the plan"), and multiple agents can run in parallel or chained to handle one big task.

The easiest way to picture it: 45 agents is like a miniature software company. On the engineering side you have someone who plans, someone who codes, someone who tests, someone who reviews, someone who handles Git. On the growth side you have someone doing SEO, someone writing email, someone running campaigns. You are the coordinator; Claude Code is the workplace; the agents are the staff. (To understand where this concept comes from, read what subagents are in Claude Code.)

What sets this apart from typing a fresh prompt every time: each agent comes prepackaged with its own instructions, scope, and bundled tools, so you never have to re-describe the role from scratch. You buy once and call it forever. The two kits are split by purpose - Engineer Kit ($99) for the code side, Marketing Kit ($99) for the growth side, or the Bundle ($149) for both - so you can buy only the agent group you actually need.

17 Engineer agents - what each one does

This is the part almost no page lists in full. Below are all 17 engineer agent names (per agentkit.best/engineer, checked 08/2026), ordered by software development lifecycle stage so they are easier to remember:

The engineer roster's strength is that it covers the whole process, not just the coding step. Plenty of subagent packs out there only give you a generic "developer"; here you get a separate agent for debugging, for review, for Git - meaning responsibilities are split the way a real team splits them. That matters because each agent can carry its own "personality" and instruction set: code-reviewer is tuned to catch bugs and regressions rather than add features, while code-simplifier focuses purely on trimming without changing behavior. You re-explain context less often, and the output is more consistent than leaning on one "do-everything" Claude for every step. Details on each agent and how they work together are in the Engineer Kit breakdown.

28 Marketing agents - what each group does

The marketing group (per agentkit.best/marketing, checked 08/2026) is organized around the marketing funnel instead of a flat list: TOFU (top of funnel - attract), MOFU (middle - nurture), BOFU (bottom - close), Core (operational core), and Support (support roles). The sales page only names 6 of them explicitly; the rest are described by group:

One honest thing to know: the marketing agent count is inconsistent across AgentKit's own pages. The home page says 28 marketing agents (45 total), but the /marketing subpage shows a higher number. I am keeping the 45 framing (17+28) from the home page, but I am flagging this gap so you can double-check at purchase time - no other page points it out. To see each marketing agent in detail, read the Marketing Kit breakdown.

Want to see the full lineup for yourself? The public roster is still missing a few marketing agent names. The surest way is to check the full 45-agent roster on the official site and reconcile the numbers against the latest version.

How do agents differ from Skills / Subagents / Commands?

This is where newcomers get most tangled - often assuming "108 skills" and "45 agents" are the same thing. These are four different concepts:

• Agent - a named AI role you call by name, which can run in parallel or chained. AgentKit has 45.
• Skill - a capability Claude loads on demand (for example, a code-review skill or an SEO-writing skill). AgentKit has 108+.
• Slash command - a command you type directly to trigger an action (for example, a fixed workflow). AgentKit has 95+.
• Subagent - Claude Code's underlying concept: a secondary Claude session with its own context. The agents in the kit are built on this subagent mechanism.

Put differently: a subagent is the technical foundation, an agent is a prepackaged role running on top of it, and skills and commands are the tools the agent (or you) draw on. To grasp the underlying mechanism, see the subagents guide. For an overview of the full capability set, read the 108+ skills overview.

A real example: one feature moving through several agents

The way 45 agents pay off is not calling them one at a time - it is orchestration. A typical task I run often - adding a small feature (say, an API endpoint plus a form) - moves through an agent chain like this:

1. planner splits the work into clear steps and identifies the files to touch.
2. scout probes the existing repo to learn its patterns; scout-external looks up library docs if needed - these two can run in parallel (query fan-out) because they do not collide.
3. fullstack-developer writes the code from the plan.
4. tester writes the tests, then code-reviewer hunts for bugs - run sequentially, since review needs finished code plus tests.
5. git-manager commits by convention and opens the PR.

# illustration of calling agents by name through the ak CLI
ak agent planner "Plan adding a /orders endpoint + an order-creation form"
ak agent scout "Find the existing controller pattern in the repo"
ak agent tester "Write tests for the new /orders endpoint"

Being honest about what is smooth and what needs hands-on work: the planning and repo-probing steps run very smoothly and save real time reading code. But the actual coding step usually needs a careful look from me - the agent can misread a business constraint, and the review step sometimes misses subtle bugs. Good orchestration is knowing when to let an agent run and when to stop and check, not pressing a button and walking away.

Are 45 agents worth it? Who should and should not use them

Straight answer: worth it if you need to coordinate many kinds of work at once; a bit much if you only write the occasional script. The real value of 45 agents is not "having 45" - it is that responsibilities are split cleanly and the agents coordinate with each other.

• Should use: solo builders/indie hackers doing both the product and go-to-market (they need engineer and marketing); small teams wanting to standardize their dev process; anyone already using Claude Code daily who wants to level up into "an AI team."
• Not yet: people just learning Claude Code (get the basic skills/subagents down first); anyone writing one-off scripts; anyone who does no marketing, since the marketing kit will sit idle.

The downsides worth stating plainly: almost nobody uses all 45 agents; a few have overlapping roles (the line between researcher, scout, and scout-external is sometimes blurry). And as noted, the marketing numbers contradict themselves - verify live before you decide. Weigh the cost more carefully in the AgentKit pricing breakdown, or read the full AgentKit review for the complete picture.

If you have decided you need the whole lineup, you can get AgentKit (20% off via link) directly from the official site (kits $99, bundle $149, no recurring fee listed; there is a money-back guarantee and lifetime updates on the kits).

Frequently asked questions (FAQ)

How many agents does AgentKit have?

Per the agentkit.best home page, AgentKit has 45 AI agents: 17 engineer + 28 marketing. Note that the marketing subpage sometimes shows a higher number, so reconcile the count live before buying.

How do you call an agent?

You call an agent by name through the ak CLI (for example, ak agent planner "..."), right inside your Claude Code environment. Many agents can run in parallel or chained.

Is this AgentKit different from OpenAI AgentKit and agentskit.co?

Yes, completely different. This post is about AgentKit for Claude Code at agentkit.best. OpenAI AgentKit (launched 10/06/2025) is a separate OpenAI product; agentskit.co is a different kit from a different developer.

Can agents run in parallel on their own?

Yes. Agents that do not collide (for example, probing the repo and looking up external docs) run in parallel; steps that depend on each other (like review after code exists) run in sequence.

Do I need Claude Code Pro?

AgentKit runs on Claude Code (and Codex/Copilot). You need a working Claude Code environment; the right Claude Code plan depends on your usage - see the article on Claude Code pricing for more.

How much is it, and where do I buy?

Engineer Kit $99, Marketing Kit $99, both as a Bundle for $149 (no recurring fee listed). Buy at agentkit.best; there is a money-back guarantee and lifetime updates on the kits.

Conclusion & links

AgentKit's 45 agents are an "AI team" with clean division of labor: 17 engineer agents cover the whole code lifecycle, 28 marketing agents cover the whole growth funnel - called by name, run in parallel or chained. The value is in the orchestration, not the count, and you will almost never use all 45. Remember to verify the numbers live (marketing wavers between 28 and 32) and think hard about whether you truly need both kits. Read the full AgentKit review before you decide.