Skip to main content

​
Overview

An agent is an AI worker in WorkLLM that completes a task for you from start to finish. You describe what you want done. WorkLLM asks a few questions, plans the steps, and then carries them out on its own. It uses your knowledge, your connected apps, and the more than 200 AI models in your workspace. You do not need to retype prompts or add context each time. This article explains how agents work and how you can start using them.

This article explains the following topics.
  • What an agent is, with examples.
  • How Agent 2.0 works: describe the task, answer a few questions, review the steps, and let it run on its own.
  • What agents can and cannot do, and how they start.
  • The three ways an agent can work.
  • How agents differ from technical workflow tools such as n8n and Make.
  • Use cases and examples.
  • How agents use credits, and how your data is protected.

​
What is an agent

An AI agent is an AI Coworker which takes care of the full end-to-end workflow for a user. An agent is built around one specific task or use case. A chat is open-ended, so you can ask it anything. An agent is purpose-built. It knows what it is for, what material to use, and what a good result looks like. Agents can also run a whole piece of work from start to finish. These are the kinds of jobs agencies and consulting firms hand to them.
  • Weekly client report. Every Monday, the agent pulls last week’s Google Ads and Meta Ads results into a report and slides, and emails them to the account lead.
  • Competitor watch. Every weekday morning, the agent checks competitor websites, pricing pages, and launches, and sends a short digest to Slack or email.
  • Content brief. Given a topic, the agent researches keywords, looks at the top-ranking content, finds the gaps, and writes a brief your writers can use straight away.
  • Client session follow-up. After each session, the agent drafts the follow-up message and a worksheet for the client.
  • Weekly status report. Every Friday, the agent pulls progress from your project tools and Slack threads and writes progress, blockers, and next steps for each team.

​
How Agent 2.0 works at a glance

Building an agent takes four steps.
  1. Describe the task. Write what you want in plain language, the way you would brief a new colleague.
  2. Answer a few questions. If something is missing or unclear, WorkLLM asks clarifying questions before it builds the plan, such as which apps to use or what should start the agent.
  3. Review the steps. WorkLLM generates a step-by-step plan. The plan shows each step, the apps it connects to, the knowledge it uses, and the full instructions. You can edit every step, so nothing runs as a black box.
  4. Let it run on its own. Save the plan as an agent. It can then run on a schedule, or when something happens in one of your connected apps.
You do not need to learn prompt engineering or wire steps together yourself. Agent or chat prompt. You can also run a multi-step task from a single prompt in chat, for example researching a topic, drafting a post, and emailing it. That is a one-time run that you start yourself. Create an agent when you want the same work to be repeatable, shareable with your team, or automatic.

​
What agents can and cannot do

Agents can do the following.
  • Answer questions from your own documents, past work, and connected apps, and name the source. A new hire can ask the leave policy, or a consultant can ask what a client’s engagement covers, without anyone searching through folders.
  • Build the plan from your description and ask what they need to know, such as your tools and what should start the agent.
  • Run on a schedule or on a trigger, for example a competitor check every weekday morning, or a triage pass whenever a new email arrives.
  • Work with the apps you connect, such as Google Workspace, Slack, Jira, HubSpot, Notion, and Salesforce, and with external MCP servers.
  • Use organization memory when you point an agent to it, so your products, tone, and client details do not need re-explaining every time. You can also run an agent without any memory context to keep a task separate.
  • Live in your shared workspace, so teammates can use and build on them.
  • Keep a version history when you update an agent.
  • Show usage reports: who used an agent, how often, and the credits used.
An agent can start on a schedule, when something happens in a connected app, or on demand. Any action in a connected app can be the trigger.  Agents cannot do the following.
  • Operate tools through your browser, such as clicking through LinkedIn or a CRM.

​
Types of agents

Every agent works the way your task needs it to. It can answer questions on demand, complete a task in one click, or run entirely on its own. Get instant answers. The agent answers from your company’s documents, past work, and connected apps, so nobody has to search through files or ask a teammate. It suits product Q&A, policy questions, client context, competitive research, and onboarding. Complete a task in one click. Someone fills in a few details and gets a finished draft, summary, or report back. Anyone can run it, with no AI experience needed. It suits writing, summarizing, drafting, researching, reporting, and any task your team does more than once a week. Run automatically. You set a schedule, an app event, or a trigger, and the agent runs in the background from then on. No one has to remember to start it. It suits multi-step automations, scheduled reports, and event-triggered tasks.

​
How agents differ from n8n, Make & Zapier

Tools like n8n, Make, and Zapier are powerful and flexible. You build a workflow on a canvas and set up each step yourself. That gives technical teams fine control over how each connection behaves. WorkLLM agents start from a description instead. You explain the task, WorkLLM builds the steps, and you review and edit them. The two approaches suit different needs. If you need to wire up custom connections in detail, a canvas tool may be the better fit. If you want to describe the work and keep control through review, Agent 2.0 is built for that.

​
Who agents are for

Agents suit any work that repeats. Agencies and consulting firms are a natural fit, and the same patterns apply to operations, support, and HR teams.

Use cases and examples You can also start from a ready-made agent and adjust it. For agencies and consulting firms, WorkLLM’s ready-made library includes the Proposal & BD Agent, New Business Agent, Frameworks & Firm Knowledge Agent, Engagement Context Agent, Brand & Client Context Agent, and Competitive Intelligence Agent. Change the inputs, output format, tone, or data source, or build your own from scratch.

​
Agents and credits

Agents use credits, the same unit that measures all AI work in WorkLLM. In brief:
  • Creating an agent, testing it, and running it all use credits. What a run costs depends on the model each step uses, how much text it reads and writes, and the type of work involved, such as a web search. 
  • Connecting an app costs 10 credits (only once). 
  • If you pick a model that runs on your own API key (BYOK), the model itself does not use WorkLLM credits. Features such as memory and web search still do.
  • WorkLLM cannot show the exact cost before a run, because it depends on how much the model writes.
For typical costs, how to see your usage, and ways to make your credits go further, read How WorkLLM credits work.

​
Agents and data security

  • Zero data retention. All models on WorkLLM follow zero data retention, so model providers do not keep your data.
  • No training on your content. Content you upload is never used to train models, and you can delete it permanently.
  • Role-based access controls. Admins control who can run agents, who can build them, and who can access which data. Nothing is open by default.
  • Full audit logs. Every agent run is logged. Admins can see who ran what, when, and what the output was.
For the full details, see the security documentation.

​
Frequently asked questions

  • Do I need coding or prompt-writing skills?
    No. You describe what you want in plain language, and WorkLLM builds the steps.
  • What if my instructions are unclear or incomplete?
    WorkLLM asks clarifying questions before it finalizes the plan.
  • Can I edit the steps before an agent runs?
    Yes. You can review and edit every step, including the apps, the knowledge, and the instructions.
  • Can an agent run on a schedule?
    Yes. Agents can run on a schedule or when an event happens, such as a competitor check every weekday morning. 
  • Can I share an agent with my team?
    Yes. Agents live in your shared workspace, so teammates can use and build on them.
  • Can I see what changes when someone updates an agent?
    Yes. When you update an agent, the version history shows the new and old documents by name.
  • Can an agent use my other apps?
    Yes, with the apps you connect. Each new connection costs 10 credits once. Check the integrations page for the current list of supported apps.
  • Do agents use my credits if I bring my own API key?
    The model itself does not use WorkLLM credits when you pick a BYOK model. Memory, web search, and connecting apps still use credits. WorkLLM never switches between your key and your credits on its own.
  • Is my data used to train AI models?
    No. Your content is never used for training.