Work
Real systems, running in production. We start with our own companies, then install the same patterns for clients. Third-party clients are described by type, not by name.
Case studies
Three different businesses, one pattern: the AI does the routine work, gathers the details, and hands anything binding to a person for approval.
Our own operations: an AI executive assistant that runs the front office
SynthOps and its sister companies (we are client zero)
Situation
- One founder running several businesses, with email, texts, Slack, and calls spread across a dozen accounts
- Important messages buried under routine ones; follow-ups slipping
- No assistant could see the whole picture, so everything routed back to the founder
What we built
- One searchable record of email, texts, and Slack across every business
- An AI assistant, Sonya, who answers our phone line, takes messages, and posts a brief to the right channel
- Outbound calls on request: ask in Slack, she places the call and reports back in the same thread
- Drafted replies and reminders, sent only through an explicit outreach switch and only to people who expect them
In production
- In daily production since early 2026
- The phone number on this site is answered by her
- Runs on a small office computer plus a handful of low-cost cloud servers
Stack / surface area: Phone (Twilio) · realtime voice AI · email/text/Slack sync · approval queue · audit log
Wellness center: an AI marketing assistant that can’t press send
A multi-service wellness center
Situation
- The owner was paying for an all-in-one marketing platform and still doing most of the work by hand
- Campaigns, contact lists, and promo memberships lived in different places
- Keeping text-message consent and contact lists clean was a manual chore
What we built
- A member portal on their own database, replacing the old platform
- An AI assistant that drafts campaigns, manages contact groups, and creates or pauses promo memberships
- A send-proof design: the assistant has no tool that delivers a message. It writes drafts for the owner to review
- Every change requires a one-time confirmation from the owner, in a later message. It cannot approve itself
In production
- Campaigns arrive as reviewable drafts instead of blank pages
- Social media inquiries land in the same contact record as everything else
- Consent and list hygiene built into the workflow, not left to memory
Stack / surface area: Member portal · AI assistant · Instagram/ManyChat intake · confirmation contract
TV graphics department: an AI coordinator and builder for a network series
A screen-graphics company working on network and streaming TV series
Situation
- Dozens of on-screen graphics per episode, each with revisions, references, and deadlines
- Notes and files scattered across email, Slack, and cloud storage
- Years of past production art that nobody could find when it was needed
What we built
- An AI show coordinator in Slack who researches across email, files, and the production board and answers in the thread
- An AI graphics builder: request a graphic in Slack, it gathers references, builds it, tests its own output, files it, and posts a review render
- An AI archivist that catalogs production art with searchable descriptions
- A hard rule: nothing is delivered to the studio without a human approving it
In production
- Every graphic for an episode tracked in one place, with status and links
- Builds arrive with the evidence attached, ready for a person to approve or give notes
- Past work is searchable by what it shows, not by folder name
Stack / surface area: Slack · cloud file storage · AI coordinator · graphics build worker · searchable archive
What this proves
We’re not selling “automation.” We’re building systems that stay useful when reality deviates.
We don’t automate the happy path only.
The “human-feeling” part is the system knowing when it’s outside the happy path—then handing off with a complete brief and safety checks.
- Edge-case inventory + decision points (no mystery work)
- Safe fallbacks and approvals (don’t guess under uncertainty)
- Human handoff with a complete brief (no “can you send more details?” loops)
We ship working increments.
We design around operability: logs, metrics, and runbooks. Then we iterate based on what edge cases are actually happening.
- Clear KPI baselines (then improve with iteration)
- Traceability: “why did it take this path?”
- Incremental rollouts with rollback paths
Capabilities (selected)
Concrete building blocks we combine to make handoffs humane and automation operable.
Who-owns-what + clean handoffs
Complete-brief templates
Form/email/chat intake triage
Approval steps + safe fallbacks
Audit logs
Dashboards + alerts
Internal tools (admin panels)
Knowledge workflows
Integrations (CRM/ticketing/data)
Custom apps (web/mobile/desktop)
Event-driven workflows (webhooks/queues)
Runbooks + escalation playbooks
Permissioning + least privilege
Measurement plans + KPI baselines
Want something like this for your team?
We’ll find the breakpoints and fix the handoff—fast. Start with a fit call and we’ll propose a safe first slice to ship.