Claude Cowork's Usage Data Shows Sales Is Falling Behind on Agentic AI

The number that stopped me

I read a lot of vendor-sponsored AI SDR data, so I pay closer attention when a lab reports on its own product's actual usage instead of a case study picked to sell licenses. Anthropic did exactly that this month with Claude Cowork, its general-purpose agent for non-coding work. The company pulled 1.2 million anonymized Cowork sessions from more than 600,000 organizations over the last two weeks of May 2026, and the breakdown got covered by SiliconANGLE and TechCrunch when Anthropic expanded Cowork to web and mobile. Sales and revenue operations came in at roughly 4% of usage. Business process operating, things like consolidating updates into a report or reconciling spreadsheets across finance and HR, was the single largest category at 33.4%. That is an 8x gap, inside the same product, in the same two weeks.

What Anthropic actually published

Cowork is Anthropic's answer to the question of what an agentic tool looks like once you take the coding assumption out of it. It launched as a desktop app and expanded to web and mobile so a task can start at a desk and finish on a phone. The headline finding both outlets led with is that software development, the most publicized use case for agentic AI, made up only 8.7% of Cowork sessions. That is a smaller share than business process operations and smaller than content creation and copywriting too. The company's framing, per TechCrunch's reporting, is that "everyday business work is on the rise" as the real center of gravity for agentic tools, not code.

The full breakdown by category

Here is the category split as reported, ranked by share of sessions. I am only including the figures both outlets confirmed.

Business operations and content creation combined already account for close to half of all Cowork sessions. Sales revenue operations sits near the bottom of the named categories, just above document extraction.

Why sales is the smallest slice

I do not think this means sales teams are less capable of using AI agents. I think it means sales already has a dense, mature category of point tools, sequencers, dialers, conversation intelligence, CRM automations, that quietly absorbed a lot of the workflow before a general agent like Cowork ever showed up. Finance and ops teams reconciling spreadsheets by hand had no equivalent packaged tool doing that specific job for them, so a general agent was the first real upgrade they had seen in years. A rep already living inside Salesforge or a CRM sequence has less obvious white space for a general-purpose agent to fill, at least for the parts of the job that already got automated a product cycle ago.

The difference between a coding agent and a general agent

It is worth being precise about what Cowork is, because I see people conflate it with Claude Code. Claude Code is built for writing and shipping software, the tool I use to build custom GTM tooling, sync data between systems, and script one-off reporting. Cowork is explicitly the non-coding sibling, aimed at the "work around the work" that keeps a company running but rarely shows up in anyone's job title. That distinction matters for where I think GTM teams should actually be looking. Coding agents are already reshaping how GTM engineers build tooling, a trend I covered in my piece on the rise of the GTM engineer. Cowork's data says the bigger untapped opportunity right now might be the boring administrative half of a revenue team's week, not the custom scripts.

What "business process operating" has in common with GTM work

Look past the category label and a lot of what counts as "business process operating" in Anthropic's data is functionally identical to work a revenue team does every week. Reconciling spreadsheets across departments is not far from reconciling pipeline numbers between a CRM and a spreadsheet a VP actually reads. Building an onboarding checklist is the same shape of task as building a new rep's 30-60-90 ramp checklist. Pulling scattered updates into one report is a weekly QBR prep task by another name. None of that is sales-specific enough to need a dedicated AI SDR product. It is exactly the kind of general, repetitive, text-heavy task a general agent like Cowork is built for, and it is very likely already showing up inside that 33.4%, just uncredited to sales because the task itself sits in the operations category.

Three sales workflows that fit the pattern

If Cowork's biggest category is process and reporting work, here are three GTM tasks I think map directly onto it and are worth handing to a general agent before you reach for a dedicated point tool.

1. Weekly pipeline reporting. Pulling stage changes, slipped deals, and rep activity into one summary for a leadership review is pure business process operating with a sales label on it.
2. New rep onboarding checklists. Building and updating a ramp plan, tooling access list, and first-30-days checklist is close to identical to the onboarding-checklist example Anthropic cited directly.
3. Account research briefs. Consolidating funding news, hiring signals, and tech stack changes into a one-page account brief before a call sits right at the intersection of the research and business-process categories.

Why I do not read this as sales being behind forever

I would be careful reading a two-week snapshot as a permanent verdict on sales and AI agents. This is one product, one 14-day window, and Cowork itself is new to web and mobile as of this data. What it does tell me is that the sales function has not yet built the same habit of reaching for a general agent for its administrative load that finance and ops teams apparently have. That is a gap I expect to close, not a ceiling. The categories most primed for growth are the ones that are process-shaped rather than selling-shaped, because that is where a general agent adds obvious leverage without touching the parts of the job, live prospect conversations, deal judgment, that still need a human or a purpose-built AI SDR.

How this shows up in my own stack

My own stack splits the same way this data suggests it should. Salesforge handles sending and sequencing and Leadsforge handles lead data, both mature, purpose-built categories where I want a vendor's years of hardening, not a general agent improvising. Agent Frank runs the AI SDR motion end to end for the same reason. Where I reach for a general-purpose or coding agent is the process layer around that stack, pulling send data out of Salesforge into a report a client actually reads, building onboarding docs for a new hire, drafting account briefs before a call. That is exactly the "business process operating" bucket Anthropic's data flags as the biggest category, and it is the part of my week that a dedicated sales tool was never going to automate for me.

What to actually do with this data

Do not read this as an argument to hand your reps' selling motion over to a general agent. Read it as a checklist of the reporting, onboarding, and research busywork sitting on your team's plate that has a much better chance of disappearing into an agent session than most people assume. Start with whatever recurring report currently takes a rep or manager more than 30 minutes to assemble by hand. If it is mostly pulling numbers or notes from three places into one document, that task is a near-perfect match for what Anthropic's own users are already doing with Cowork.

FAQ