AI agents on the Scalekit network are doing real work in real systems: half of their calls land in CRMs, notes and docs, and most belong to multi-step jobs. This first edition shows where that work happens and which apps agents are adopting fastest.
CRMs and notes
top riser growth
multi-step jobs
Each call is an agent using a real app on someone’s behalf, such as reading a pipeline or filing a ticket.
of jobs chain two or more tool calls
Most of it is pipeline reads, deal updates, lead enrichment and ticket triage. Productivity tools take another 25%+, so the two together are over half the network.
of agent calls are sales, marketing and support work
An agent usually reads what the job needs and writes once at the end, for example a CRM record or a sent message. Writes should grow as teams let agents act without a human sign-off.
of agent calls write back to a real system
A typical job reads a ticket, pulls the account from the CRM, checks the docs and posts an update. Most of these chains repeat a job the agent has run before.
of jobs run six or more calls across apps
These are the clients that will connect to your agent or MCP server once it’s live.
What it meansClaude clients make 40%+ of calls, and in-house agents make 30%+. A lot of teams build their own agents as well as using Claude or ChatGPT.
What it meansGTM and Productivity each take 25%+ of calls, half of the network between them. People Ops is small for now, but it’s growing fast.
Every app on this list saw agent calls grow by at least 30%. If you’re deciding what to connect next, this is a good place to start.
What it meansMeeting-notes tools and newer CRMs grew fastest. Granola, Fathom and Attio each more than doubled their agent calls.
Every call is one of five things: read, create, update, delete or execute.
Create means a new record or a sent message. Execute means running a job, sync or enrichment that doesn’t change a record directly.
What it meansAbout 6% of calls are writes, and half of those create something new, such as a record or a message. GTM writes back most, at about 10% of its calls. Many writes still need a person to approve them, so this share should grow as more agents reach production.
What it means20%+ of jobs run six or more calls across apps. Most are repeats of a job the agent has run before.
Pick an app to see what agents usually use with it, and why.
What it meansMost of these jobs pair the app where the record lives with one specialist tool, like a CRM with an enrichment app.
Agents do not browse. They arrive with a job, read everything the job needs, and write once at the end.Agents on the Clock · September 2026
This report is built from aggregate call counts. We never read the contents of a call, so no payloads, prompts or customer data are included. All figures come from a sample of September 2026 traffic and are rounded down. Monthly totals are a 30-day run rate. Per-app growth and pairing figures are directional. They describe the sampled traffic, not the whole market.
Every call runs with the end user’s own access, so your agent acts as the person who asked.
// connect your agent to its tools
npx @scalekit-inc/cli setup 
We use cookies, so things load fast, we learn what to fix, and you can always reach us on chat
