What is Cowork and why does it matter for procurement?
Most AI tools are chat boxes. You talk, they respond, you copy-paste the output somewhere useful. Cowork is different in a fundamental way: it acts. It sits inside the same Claude desktop app as your chat window, one click away, and instead of responding to your prompt it executes it. It reads your files, writes to your folders, opens your browser, clicks buttons, fills forms, runs scheduled jobs. Unattended.
Anthropic built Cowork on the same architecture that powers Claude Code, their developer tool. They wrapped it in a UI that anyone can use. No terminal, no Python, no setup beyond downloading the desktop app. It is currently in research preview, available on all paid plans from Pro upwards, on Mac and Windows.
Procurement is one of the best-fit functions for this kind of tool. Our work runs on repetition: RFP drafts, supplier scorecards, spend reports, PR form fills, contract reviews, category summaries. It is exactly the kind of high-volume knowledge work where an autonomous agent delivers a strong return compared to a conversational assistant.
The alternative paths (multiple vertical SaaS tools or a custom internal build) are either expensive, slow, or both. Cowork sidesteps that entirely. It is a general-purpose agent you install in fifteen minutes and then shape to your specific workflows through skills and plugins, without writing a line of code. If you want a practical guide to getting started with AI in procurement, we publish those on our resources page.
Claude Cowork features that matter for procurement
01. The desktop app
One app. Chat and Cowork in the same window. No friction.
The Claude desktop app is one of the few AI products that feels like it belongs on a professional's computer. Chat, Cowork, and Code live in a single sidebar. No context-switching between tools, no browser tab graveyard. When you are mid-conversation and realise you need to actually execute something, you flip to Cowork. When Cowork is running in the background, you can still chat in the other tab.
Start a category market analysis in Chat. When you are ready to turn it into a formatted supplier brief saved to your folder, switch to Cowork without leaving the app.
02. Browser automation via Claude in Chrome
Cowork can operate your browser. This is the part that makes people's jaws drop.
Pair Cowork with the Claude in Chrome extension and it can take actions in any website: clicking, filling, navigating, extracting, like a human using a computer. The model can see the page, understand what is on it, and interact with it. This is not pre-recorded macros. It adapts when the page changes. Anthropic explicitly advises limiting this to trusted sites and using it cautiously with sensitive data.
Imagine pointing Cowork at your Ariba or Coupa instance to pull open PRs, check approval status, or draft a requisition from a structured brief. Or opening a supplier portal and extracting pricing tables into a comparison sheet.
03. Skills
Your repetitive tasks, turned into one-click commands.
Skills are saved, repeatable task templates. You define the workflow once (instructions, folder access, output format) and from that point it is a single slash command. The real power is in multi-step workflows that chain tools together: download data from a shared folder, format it, convert it to PowerPoint slides, all in one command.
/spend-report pulls the latest data and produces a formatted PowerPoint with category slides. /supplier-scorecard takes a folder of tender responses and produces a comparison doc. /rfp-first-draft takes a SOW and generates a structured RFP in your template.
04. Plugins
Pre-built specialist bundles for your role, your team, your tools.
If Skills are custom automations you build yourself, Plugins are pre-packaged suites that bundle skills, connectors, and sub-agents into a single install. A plugin wires together multiple tools and gives Cowork specialist context about your function from day one.
A "Procurement Specialist" plugin could bundle an RFP generation skill, a supplier outreach connector, an Ariba read connector, and a spend analysis sub-agent, all installed across your team in one click. We are actively building this at Moleculeone. Reach out if you want early access.
05. Private marketplace
Your team's institutional knowledge, packaged and shared at the click of a button.
Team and Enterprise admins can create a private plugin marketplace: a curated internal catalog of approved skills, plugins, and connectors specific to your organisation. You build a /spend-report skill, validate it, and push it to every procurement user in your org. Admins control what is auto-installed, what is available on request, and what is hidden entirely.
Your category manager builds a best-in-class supplier negotiation brief skill. Instead of it living on their laptop, it goes into the company marketplace, available to every buyer on the team with one click, version-controlled, and improvable over time.
06. Scheduled tasks
That report you built a skill for? It now runs itself.
Type /schedule in any Cowork task and you can set it to run daily, weekly, monthly, whatever cadence you need. Cowork executes on its own as long as your computer is on. No orchestration tools. No Zapier. No code. Just Cowork and a schedule.
Weekly supplier price variance report every Monday at 7am. Monthly tail spend summary from your ERP export every first of the month. Automated contract expiry alerts from your contracts folder every Friday.
07. Model capability
1 million token context. This matters more than you think.
Cowork runs on Claude Sonnet 4.6, with a 1M token context window in beta. That means you can feed it an entire contract repository, a year's worth of supplier communications, or a 500-page spend analysis and have it reason across the whole thing in one session.
Upload 12 months of supplier invoices and ask Cowork to identify price creep, anomalies, and missed rebate triggers. Or feed it every tender response and ask for a comparative evaluation matrix with scoring rationale.
The part most people miss: it is not just Cowork
Claude is not just a desktop agent. It also lives inside Excel and PowerPoint. When you combine those add-ins with Cowork, you get something closer to a unified AI workspace than anything else currently available for procurement teams at this price point.
Claude in Excel
Your spend data, analysis-ready without leaving the spreadsheet.
The Claude for Excel add-in brings conversational AI directly into your spreadsheets. Ask it to analyse spend by category, identify outliers, build pivot tables, apply conditional formatting, or write formulas in plain English. It runs on Opus 4.6 and supports native Excel operations including pivot table editing. The interesting bit for procurement: it shares context with the PowerPoint add-in, so Claude can analyse data in Excel and move a chart directly into a presentation without you manually exporting anything.
Procurement workflow: Drop your monthly spend extract into Excel. Ask Claude to categorise by supplier tier, flag anything above contract thresholds, and build a summary pivot. Then push the key charts into this month's category review deck in PowerPoint automatically.
Claude in PowerPoint
From data and brief to polished deck, without copy-pasting.
The PowerPoint add-in lets Claude read your current deck, understand the context, and create, edit, or add slides based on instructions. Combined with Excel context-sharing, it takes the output of a spend analysis and builds slides around it with minimal manual effort. You can also add Skills to the add-in, extending it with procurement-specific templates the same way you customise Cowork.
Procurement workflow: You have built a supplier benchmark analysis in Excel. Claude in PowerPoint reads the data, understands your deck's existing structure, and produces three new slides: an executive summary, a supplier comparison table, and a recommended action slide, formatted consistently with the rest of your presentation.
The enterprise bundle: everything in one subscription
On Anthropic's Enterprise plan, you do not need to assemble a procurement AI stack piecemeal. You get all of this under a single subscription, and that is unusual. Most vendors would charge separately for each of these surfaces.
Based on what we have seen work with early clients: start with Cowork and Skills, build two or three high-frequency task automations in the first month, then layer in the Excel and PowerPoint add-ins. The value of these tools working together is much higher than any of them used in isolation, and the whole thing costs a fraction of what most procurement teams spend on point solutions that do far less.
What does this actually cost for a team of 10?
Pricing is one of the strongest parts of the Cowork story for procurement teams, especially compared to the alternatives. Here is a realistic picture for a 10-person team, with an honest note on which plan we would actually recommend based on real usage.
| Plan | Per seat / mo | 10 seats / mo | Annual |
|---|---|---|---|
| Team Standard Central billing, admin controls, plugin marketplace. Pro-level usage per seat. |
$25 | $250 | $3,000 |
| Team Premium (our pick) Same as Standard plus Max-level usage. Needed for regular Cowork users. Includes Claude Code. |
$100 | $1,000 | $12,000 |
| Enterprise SSO, audit logs, SCIM, compliance API. Min. 20 seats. Usage billed separately at API rates with no seat-level caps. |
Contact sales | — | Custom |
| Our recommended starting point: 10 x Team Premium, annual | $1,000 / mo | $12,000 / yr | |
Pricing as of 17 March 2026. Always check claude.com/pricing for official, up-to-date pricing.
To put $12,000 per year in context: that is roughly what many teams spend per quarter on a single-function SaaS procurement tool, before implementation or integration costs. For that you are getting a general-purpose agent, browser automation, Excel and PowerPoint AI, Claude Code, and a private skills marketplace for a team of 10.
A practical note on Standard versus Premium: Standard is a reasonable starting point to evaluate Cowork before committing. But if your team plans to run Cowork seriously (multi-step tasks, scheduled automations, complex document processing), you will hit limits on Standard seats and it will frustrate people. Pilot with a few Premium seats first, validate the return, then roll out.
On Enterprise: the usage-based model means costs scale with actual consumption rather than being capped. For teams running heavy automated workloads this could work out better than paying per seat for headroom nobody is using, but it requires modelling your expected token usage before committing. We are happy to help teams think through that calculation.
This is not only a procurement story
Our focus is procurement. That is where we spend our time and where we can speak with authority. But we would be doing you a disservice if we did not point out the obvious: the knowledge work that Cowork is good at is not unique to procurement. Finance teams running month-end reports, marketing teams producing campaign briefs and performance decks, legal teams reviewing and summarising contracts, HR teams synthesising candidate feedback. They are all doing the same kind of high-volume, file-heavy, repetitive work where Cowork delivers.
This matters for one practical reason. It is much easier to get an AI platform approved and budgeted at the company level than the department level. If procurement leads the internal case for Claude but frames it as a company-wide tool that finance, legal, and marketing will also benefit from, the business case gets a lot stronger and the per-team cost drops considerably.
If you are building an internal business case and need help framing the cross-functional angle, that is something we can help with through our AI procurement consulting services, even though those teams are not our primary focus.
Claude Cowork limitations: what we don't like after 60 days
We would rather be useful to you than write a glowing product review. Here is what frustrates us after 60 days.
Chat and Cowork share the same usage pool.
Cowork tasks are compute-intensive. A complex multi-step task can consume far more of your usage allocation than a standard chat message. The problem is that your pooled credits are shared across chat, Cowork, and Code, so heavy chat users are eating into high-return automation runs. Our recommendation: if your team is a serious Cowork shop, route simple Q&A to cheaper alternatives and save Cowork credits for actual task execution.
Cowork stops when your computer does. This will catch you out.
Cowork runs locally on your machine. The moment you close your laptop lid, sleep your computer, or shut the app, any running task stops. For scheduled tasks meant to run overnight or early morning, your machine needs to stay on and the app needs to stay open. It is a habit adjustment that is very doable, but real. What we would like to see is a cloud execution mode where tasks run independently of the desktop. The Microsoft Copilot Cowork integration announced this month does exactly this, which gives us confidence that Anthropic's standalone product will move in this direction too.
95% seamless. 5% unexplained weirdness.
The vast majority of sessions are smooth. But there are moments where context drops mid-session, skills do not fire on first trigger, or scheduled tasks need a nudge. These moments remind you this is not production software yet. Anthropic is iterating fast. A scheduling bug that broke tasks for users last week was fixed within 24 hours. But if you are in a zero-tolerance-for-errors environment, plan for occasional friction.
Your Cowork history lives on your machine. Plan accordingly.
Cowork stores session history locally on your device, not on Anthropic's servers. Good for privacy, since your procurement data stays off Anthropic's servers. But if you reinstall the app or switch machines, that history is gone. There is also no admin-accessible audit log during the research preview, which enterprise compliance teams should flag before deploying Cowork on regulated workloads.
You cannot check in on running tasks from your phone.
The scenario we keep wanting: kick off a contract review in the morning, step away to a meeting, and approve a mid-task permission prompt from the phone during the commute. That is not possible today. Even a read-only task status view on mobile would be a big help.
External connectors for Gmail and Google Drive can be unreliable. The Chrome integration is generally more stable. Windows arm64 is not currently supported. Cowork cannot be used inside Claude Projects yet. Prompt injection risk is real when Cowork is browsing the web or reading external files. For regulated procurement workflows touching personal data, financial close, or compliance processes, hold off until audit logging is added post-research preview.
Why early adopters should get in now anyway
Those are real limitations. So why are we still recommending it?
Because the pace of development since January has been unlike anything we have seen from an AI product at this stage. Cowork launched for Mac in mid-January. Windows followed in February with full feature parity. Enterprise connectors for Google Drive, Gmail, DocuSign, and FactSet shipped in late February. Scheduled tasks were added. A plugin marketplace with admin controls launched for Team and Enterprise. The 1M token context window arrived in beta. And in March, Microsoft built an enterprise cloud version of the same Cowork engine into M365, which tells you a lot about where the underlying architecture is headed.
Anthropic has not published a specific general availability date for standalone Cowork. Based on the trajectory (multiple major features in the first ten weeks, a $30 billion cloud infrastructure deal with Microsoft signed in late 2025), our estimate is that a more capable and stable production version is likely within the next two to three quarters. Teams that are building their skill libraries and plugin configurations now will have a real head start when it lands.
Claude Cowork verdict: is it right for your procurement team?
We are a new consultancy with a small client base and no special relationship with Anthropic. We can only speak from what we have seen in the field. And what we have seen is this: procurement teams that start building with Cowork today, even with its current limitations, are developing an operational muscle that is hard to replicate quickly. The skills, the plugin configurations, the accumulated knowledge of which workflows are worth automating and how to set them up well: that builds over months, not days.
The combination of Cowork, Claude in Excel, Claude in PowerPoint, and Claude Code in a single subscription is the most practical AI stack we have come across for procurement teams who want to move beyond chat and start actually automating work, without a large SaaS contract or an 18-month engineering engagement.
When browser automation for Ariba, Coupa, and SAP is reliable and auditable, this product will be disruptive in our industry. We are helping our early clients get in position ahead of that. If you are thinking about the same move, we are happy to share what we have learned.
What we love
- Unified chat, Cowork and Code in one app
- Browser automation for any web-based tool
- One-click skills for repetitive procurement tasks
- Private plugin marketplace to share approved skills across the team
- Scheduled tasks that run on their own
- 1M token context for large document sets
- Claude in Excel and PowerPoint with shared context
- Full bundle at $100 per seat per month on Team Premium
What needs work
- Laptop must stay open for tasks to run
- Shared usage pool eats Cowork credits fast
- Research preview reliability — around 5% friction
- Local history only, no cloud sync or backup
- No mobile view for running task status
- No audit logs yet for compliance teams
- Connector reliability for Gmail and Google Drive
Frequently asked questions about Claude Cowork for procurement
What is Claude Cowork and how does it work for procurement teams?
Claude Cowork is Anthropic's desktop agent mode, available inside the Claude app on Mac and Windows. Unlike a chat tool, Cowork executes multi-step tasks autonomously. It reads your files, writes to folders, controls your browser, fills forms, and runs scheduled jobs without you watching. For procurement teams, this means high-volume repetitive work like RFP drafting, supplier scorecards, spend reports, and contract reviews can be handed off to an agent that completes them unattended, often while you do other work.
How much does Claude Cowork cost for a procurement team of 10?
Our recommended starting point is the Team Premium plan at $100 per seat per month ($1,000 per month or $12,000 per year billed annually for 10 seats). This includes Cowork, Claude Chat, Claude in Excel, Claude in PowerPoint, and Claude Code. Team Standard at $25 per seat is available but regular Cowork users will hit usage limits on complex tasks. Enterprise is custom-priced with a 20-seat minimum. Always verify current pricing at claude.com/pricing.
Can Claude Cowork integrate with Ariba, Coupa, or SAP?
Not via a certified native connector today. However, Claude Cowork paired with the Claude in Chrome browser extension can interact with web-based procurement platforms: clicking, filling, navigating, and extracting data as a human would. Anthropic advises using this cautiously on sensitive systems during the research preview, and it is not yet auditable. Full, reliable browser automation for enterprise procurement platforms is expected post-general availability.
What are the biggest limitations of Claude Cowork right now?
The five that matter most for procurement teams: (1) the desktop app must stay open (tasks stop if your computer sleeps); (2) no cloud execution or mobile monitoring; (3) no admin audit log, making it unsuitable for regulated compliance workflows; (4) Chat and Cowork share the same usage pool, so heavy chat users eat into automation credits; (5) Gmail and Google Drive connectors can be unreliable. Anthropic is shipping fixes fast, and most of these are expected to be resolved before general availability.
Is Claude Cowork better than dedicated procurement SaaS tools?
They serve different purposes. Platforms like Coupa or Ivalua are systems of record with deep ERP integrations, approval workflows, and compliance infrastructure. Claude Cowork automates the knowledge work surrounding procurement (drafting, analysing, summarising, reporting) without a six-figure SaaS contract or an 18-month implementation. For most teams, they are complementary rather than competing. Cowork handles the cognitive load that procurement platforms do not touch.
How do I get started with Claude Cowork for my procurement team?
Download the Claude desktop app, sign up for a Team Premium plan, and install Cowork from within the app. No terminal or code required. Build your first skill around a high-frequency task: an RFP first draft, a supplier scorecard, or a weekly spend summary. Run it once, validate the output, schedule it. We recommend piloting with 2–3 Premium seats before rolling out to the full team. Our resources section has setup guides and starter skill libraries specifically for procurement teams.