All insights

Copilot Licensing · 6 min read

Cowork's New Effort Selector: You Now Decide How Hard Your Agent Works

By James Wilkinson 22 August 2026

Copilot Cowork's new effort selector lets you decide how hard an AI agent works and what it costs. What it signals about the agentic era and how to use it well.

TL;DR
  • Cowork now has five effort levels: Light, Medium, High, Extra High and Max. Medium is the default.
  • Higher effort gives the agent more room to reason, but it takes longer and burns your credit limit faster.
  • Effort sits alongside the model picker, so users have two levers: which AI does the work and how hard it works.
  • This is what agentic pricing looks like: you pay for the work performed, not a seat licence, and effort is your throttle.
  • The /cost command shows what a task has consumed, which makes effort discipline easy to build into a team.

Copilot Cowork is Microsoft’s agentic system. You describe an outcome, it builds a plan, pulls context from your emails, files, meetings and data, then works through the steps and hands you the result. It’s the clearest picture yet of where Microsoft thinks work is going: you delegate, the agent delivers. (New to Cowork or unsure how it differs from Copilot Chat? Start with Copilot Chat vs Cowork: what’s actually different? and come back.)

Microsoft has just added something to that picture. An effort selector now sits in the Cowork task box, right next to the model picker, and it lets you decide how much thinking your agent puts into a task before you press go.

This is a bigger deal than it looks, because agentic AI doesn’t come with a flat monthly price. Cowork bills by usage, in Copilot Credits, and effort is now the most direct control over that spend Microsoft has ever put in front of an end user. If your firm is moving into agents, or planning to, this feature tells you a lot about how the whole category is going to work.

What the effort selector does

When you hand work to an agent, the quality of the result depends on how much room the agent has to think: how deeply it reasons, how much of its own work it checks, how far it digs into your data. The effort selector controls exactly that.

The slider runs from Faster to Smarter, with five stops:

  • Light for simple, low-stakes tasks
  • Medium the default, a sensible balance for everyday delegated work
  • High for deeper analysis or more complex reasoning
  • Extra High when the task needs a genuinely thorough pass
  • Max for your hardest work

Microsoft’s guidance is refreshingly plain: higher effort produces more thorough results but takes longer and uses your limits faster. The tooltip on Max says it uses the most credits, can run slower and is best saved for the hardest tasks. When the vendor tells you not to run the most expensive setting by default, believe them.

If this slider looks familiar, it’s because Claude Code has had the same Faster to Smarter control for a while. That’s no coincidence: Anthropic worked with Microsoft to build Cowork, and features tend to show up in Claude before a version lands in Copilot. If you want a feel for where Microsoft’s agentic tools are heading, the Claude app is often the best preview available.

Why this matters: agents are priced like staff, not software

Here’s the shift the effort selector makes visible. Traditional software has a seat price. You pay per user per month and usage is effectively free. Agentic AI doesn’t work like that. Cowork requires a Microsoft 365 Copilot licence but includes no Cowork usage in it. The agent’s actual work is metered in Copilot Credits, and consumption scales with the job: which models run, how much context gets retrieved, how many actions the agent takes and how long it runs.

In other words, an agent is priced more like a member of staff than a piece of software. A quick job costs a little. A deep, multi-source, multi-step job costs a lot. And just as you wouldn’t put your most senior person on every trivial task, you now shouldn’t run every Cowork task at Max.

The effort selector is the first tool that lets users act on that logic directly. It’s a delegation decision: how much is this task worth?

Two levers: which agent, and how hard it works

The effort selector sits next to the model picker and they do different jobs, so it’s worth keeping them straight.

The model picker chooses which AI handles the task. Auto is the default and picks for you. The effort selector chooses how hard that AI works. The same model on Light and on Max behaves very differently: one gives a quick pass, the other works the problem properly.

For most people the right setup is simple: leave the model on Auto and use effort as your main control. Don’t think about which AI to use, think about how much the task matters. If you do want to understand the model side, our guide to the Copilot model picker covers when switching from the default earns its keep.

Matching effort to the work you delegate

The judgement to build, for yourself and your team, is this: does the task need deep reasoning, or does it just need doing?

Light or Medium territory: inbox triage, quick document edits, summarising a meeting, reformatting, routine drafting. High-volume, low-complexity delegation. Running these at Max is paying senior rates for junior work.

High, Extra High or Max territory: multi-source research, data analysis, long multi-step workflows, anything client-facing or board-facing, anything where a shallow answer costs more than the credits saved. This is where the extra reasoning earns its keep.

Watch for the failure mode nobody expects. It usually isn’t people burning credits at Max on trivia. It’s the opposite: someone runs a genuinely hard task on Medium, gets a mediocre result and concludes the agent isn’t very good. Effort cuts both ways. Part of adopting agents well is giving hard work the effort it deserves.

The habit that keeps agentic spend honest

Usage-based AI rewards firms that measure. Cowork gives you the tool for free: type /cost in any task and it shows the approximate credits that task has consumed and what’s left of your monthly limit.

So here’s the habit we’d build into any team starting with Cowork. Run tasks on Medium by default. Type /cost when they finish. Keep a simple log of the task and what it cost. Within a couple of weeks you’ll have something no forecast can give you: real numbers on which of your workflows are cheap to delegate, which are expensive and which are worth turning the effort up for. That log becomes the basis for sensible credit budgets, sensible agent use cases and sensible conversations about where automation pays. It sits naturally alongside the ten habits that keep Copilot Credits under control, which cover the briefing, scoping and admin-side controls.

That’s the wider lesson in this small slider. The agentic era isn’t just about what AI can do, it’s about learning what each piece of delegated work is worth. Firms that build that judgement early will get dramatically more from their credit spend than firms that leave every dial on default and hope.

Want to go further with Cowork and agents?

Cowork is the general-purpose end of Microsoft’s agent story. Beyond it sit purpose-built agents: grounded in your own documents and systems, running your specific workflows, governed inside your Microsoft 365 environment. Both are worth understanding, and the credit economics above apply to all of it, including the harness Copilot Studio now builds on.

That’s the work we do at FiveForward: designing and building agents for your firm, and running agent workshops for the people who will build and look after them. The Agent Journey shows where Cowork sits on that path and pricing is fixed and published. If you want to talk through where to start, book a free 30 minute consultation. No pitch, no obligation.

Sources checked

Last checked: 22 August 2026.

Related reading

More on Copilot Licensing

Copilot Licensing Copilot Cowork Costs: Ten Habits That Keep Copilot Credits Under Control Copilot Cowork now bills every task in Copilot Credits. Ten working habits that keep spend sensible, from tighter briefs to the /cost command. Copilot Explainers Copilot Chat vs Cowork: What's Actually Different? Microsoft 365 Copilot can already draft emails and manage your calendar. So what does Cowork add? The difference explained, including the agentic harness. Automation Which Copilot Studio Harness Should Your Agent Be Built On? An agent harness explained in plain English, why the GitHub Copilot harness costs more and how to decide which harness each agent build belongs on. Copilot Explainers The Copilot Model Picker Nobody Switches: When to Use Default, GPT-5.5 or Claude Opus Microsoft 365 Copilot has a model picker most teams never touch. When to use the default, GPT-5.5 or Claude Opus, and why you never leave Microsoft. Copilot Agents Microsoft Copilot Cowork: From AI Chat to AI-Supported Task Completion Microsoft Copilot Cowork moves from asking questions to delegating repeatable tasks. A guide to readiness, approval steps and where humans must decide. Automation Copilot Studio's New Engine: What the GitHub Copilot Harness Changes and What It Costs Copilot Studio's default is now the GitHub Copilot harness, the engine behind Copilot Cowork. What changes, what the credits cover and the checks before you build. Agents Microsoft Copilot agents Plain-English guidance on where Microsoft Copilot agents fit, how to govern them and when to build. Capability map The Agent Journey Five stages from using your first Copilot agent to running a team of agents, with two doors at every stage. Service Copilot Studio Agent Builds Custom agents grounded in the right knowledge and built for defined jobs, in your own tenant. Next step Pricing Fixed prices for AI agent work: Discover, fixed-price builds, agent workshops and the Embed retainer. Next step Book a free consultation A free 30-minute call about the work an agent could take off your team, and whether Discover is the right next step.

Common questions

Questions about Copilot Cowork effort selector

What are the effort levels in Copilot Cowork?
There are five, running from Faster to Smarter. Light for simple, low-stakes tasks, Medium as the default balance for everyday delegated work, High for deeper analysis, Extra High for a genuinely thorough pass and Max for your hardest work. Higher effort produces more thorough results but takes longer and uses your credit limits faster.
Does the effort selector change what Cowork costs?
Yes. Cowork is metered in Copilot Credits and consumption scales with the models used, the context retrieved, the actions taken and how long the agent runs. Higher effort gives the agent more room to reason and so uses more of your limit. Microsoft's own tooltip says Max uses the most credits and is best saved for the hardest tasks.
What is the difference between the effort selector and the model picker?
The model picker chooses which AI handles the task, with Auto as the default. The effort selector chooses how hard that AI works. The same model on Light and on Max behaves very differently. For most people the right setup is to leave the model on Auto and use effort as the main control.
How do we see what a Cowork task cost?
Type /cost in any Cowork task and it shows the approximate credits that task has consumed and what is left of your monthly limit. Running tasks on Medium by default, checking /cost when they finish and keeping a simple log gives you real numbers on which workflows are cheap to delegate and which deserve more effort.