w/01
Inbox triage
Agents that read Gmail, Slack, Linear, or Intercom — categorize, prioritize, and draft responses for human review. Never autosend until the workflow has earned trust.
Service — Workflow Automation s/03
AI agents in the tools your team already uses.
We build AI agents that plug into Gmail, Slack, Notion, HubSpot, and your CRM — and take over the repetitive parts of the work. Triage, drafting, classification, reporting, research. Your team gets hours back per week.
01 / you'll recognise this
Symptoms.
- Your team spends the first hour of every day triaging an inbox.
- Someone writes the same 20 emails, differently, every week.
- Support tickets get categorized by hand, and the categorization is inconsistent.
- Meeting notes don't get actioned because nobody has time to extract tasks from them.
- Your CRM is always slightly out of date because keeping it current is someone's least favorite task.
- You've looked at Zapier and Make and they don't understand context — they just move data.
02 / the work
What we actually do.
w/02
Meeting notes → tasks
Integrations that turn Fathom, Fireflies, or Otter transcripts into structured Jira, Linear, or Asana tickets with owners and due dates — no manual extraction.
w/03
CRM enrichment and hygiene
Agents that watch for new contacts, enrich records from public sources, and keep the CRM current without a human ever typing into it.
w/04
Ticket classification and routing
AI that understands your support taxonomy better than a keyword rule, routes to the right team, drafts the first response, and flags edge cases for humans.
w/05
Research agents
Deep-research agents that produce briefed summaries before a sales call or a hiring decision — with citations, not invented facts.
w/06
Custom MCP servers
For the tools you use that don't have off-the-shelf AI integrations, we build the MCP layer ourselves — and open-source it when the client agrees.
03 / the trade
What changes. What doesn't.
You keep
- Your team's judgment — agents draft, humans approve
- Your existing tools — Gmail stays Gmail, Slack stays Slack
- Your data boundaries — agents get scoped access, never more
- Your audit trail — every agent action is logged and reversible
You get
- Hours back per person per week, measured before and after
- Fewer context-switches for the humans
- Consistency where manual work used to be ad-hoc
- A repeatable pattern for adding the next agent, once the first one lands
04 / engagement
Where this fits.
Workflow Automation is the implementation side of the engagement ladder. The ladder itself is the same for every service we run:
- step 01 Free Snapshot
We review your site or repo and send a 1-page report with 3–5 AI integration opportunities specific to Workflow Automation. 2 business days. $0.
- step 02 Full Audit — $1,200 flat
One week deep-dive. 10–15 page prioritized roadmap with estimates, risks, and an implementation plan. Credited against the project if we work together.
- step 03 Implementation — $4K–$25K
Scoped work, fixed price, clear deliverable. Most Workflow Automation projects land between $6K and $15K.
05 / stack
What we reach for.
- Models
- Claude Opus · Claude Sonnet · Claude Haiku
- Protocol
- Model Context Protocol (MCP) — custom servers for Gmail, Slack, Linear, Notion, HubSpot, Airtable
- Orchestration
- Cloudflare Workers · Durable Objects · Queues
- Integrations
- Gmail API · Slack Bolt · Notion API · HubSpot · Zendesk
- Observability
- OpenTelemetry · structured logging · every action reversible
These are examples, not rails. We pick tools per engagement based on what already lives in your stack — we don't force a preferred tech on you.
06 / questions
Questions we hear a lot.
q/01 What's the difference between this and Zapier?
Zapier moves data between apps based on rules. Our agents make decisions, draft content, and handle ambiguity — things rules can't express. We often use Zapier as plumbing under an agent.
q/02 How do you stop agents from sending embarrassing emails?
Default posture: draft-only. Agents draft, humans approve, humans hit send. Once a specific workflow has earned trust — measured by a rolling approval rate over a set number of runs — we upgrade to autosend with human-in-the-loop on exceptions.
q/03 What about sensitive data?
Scoped access always. An agent that only needs to read one Gmail label gets access to one label, not all mail. Every agent ships with a scope document that spells out what it can and can't touch.
q/04 Can we host the agents ourselves?
Yes. Default deployment is Cloudflare Workers (edge, fast, cheap) but we've shipped agents on client infra (AWS, GCP, on-prem). The pattern is portable.
q/05 How do you measure ROI?
Before/after time audits. We sample how long the manual version of the workflow took pre-engagement and compare to post-agent. Expressed in hours per week per person — the same unit the team feels.
q/06 My CRM is a mess. Can the agent still work?
If the data is bad, the agent's output will be bad. CRM enrichment often has to happen before agents can work well with it — usually the first workflow we fix when it shows up.
Start with a Snapshot.
Two business days. No strings. A 1-page report with 3–5 AI integration opportunities specific to your situation.