SHAI
SHAI is the AI agent we built to run Twin-Ex itself — it reads the entire business (pipeline, finance, inbox, calendar, knowledge base) and drafts the email, the meeting, the deck and the daily summary, with a human approving every action.
AI agent · Operations
An AI assistant that can touch a real business is only useful if it is both deep and safe: it has to see everything to help, yet it must never quietly send an email, move money or change a record on its own. Most 'AI copilots' are one shallow chat box bolted onto an app — we needed one that actually runs the studio without becoming a liability.
A tiered agent on Claude Haiku 4.5, wired into the platform: 13 specialists covering the pipeline, finance, inbox and calendar, meetings, the knowledge base and research — reaching the business through 63 tools. Read actions run freely; anything that writes — an email, a Slack message, a calendar event, a Google Doc or a deck — is drafted for one-click human approval, and destructive actions are blocked behind dual approval. Every call is logged, every shekel is metered against a monthly ceiling, and every automation keeps a manual fallback.
Not a bolted-on chatbot — an agent that lives inside the platform
Most "copilots" are a shallow chat box glued onto an app: they can talk, but they cannot really touch the business. SHAI is built the other way around — wired directly into the Twin-Ex platform, it sees the pipeline, the finances, the inbox and the calendar, and acts from inside them. It is not an add-on; it is part of the back office.
The architecture that lets AI touch a real business — without becoming the risk
An AI agent that touches the business is only useful if it is both deep and safe. So SHAI is built in tiers: read actions — read an email, check a calendar, pull a number — run freely. Anything that writes — send an email, a Slack message, a calendar event or a Google Doc — is drafted first and waits for one-click human approval. Destructive actions are locked behind a double confirmation. Every call is logged, every shekel is metered against a monthly cap, and every automation has a manual fallback. That is how you let it see everything — without letting it send anything on its own.
That is the difference between "let AI run the business" as a slogan and something you can actually switch on: every action that changes the world passes through a human.
13 specialists, 63 tools
Under the hood SHAI is not one model but 13 specialist agents — pipeline, finance, inbox and calendar, meetings, knowledge base and research — reaching the business through 63 tools, on Claude Haiku 4.5. Each specialist owns its corner, and routing stitches them into a single answer. The result is an assistant that knows the business deeply, not a generic chat you have to re-explain everything to every time.
In practice it writes the daily briefing, drafts the lead follow-ups, surfaces the deal that stalled and prepares the meeting materials — and we approve with a click instead of typing from scratch. Two people who operate like twenty.
It writes the daily briefing, drafts the follow-ups and surfaces the stuck lead — and we approve, instead of typing.