Decide · Socialhub Harness
One governed platform for every AI you put to work.
From your Marketing, Loyalty and Risk managers to the member concierge and every behind-the-scenes helper, each AI is a structured blueprint — which model, what context, which skills, which tools, which data — plus a runtimethat executes it. You configure what each AI can do. You don't rewrite code.
Rolling out with the governed agent stack.
model
the reasoning engine behind this employee
context
the brief and business facts it starts from
skills
the playbooks it may run — win-back, cross-sell…
tools
the governed actions it may take — nothing raw
data
the members, segments and metrics it may read
always on · never configurable
authorization · spend caps · budget · kill-switch · audit
Illustrative blueprint · change what the AI does by editing the blueprint, not the codebase
The problem
A pile of AI tools isn't a team you can run.
Every AI, its own island
One vendor for the chatbot, another for the copy, a script for the fraud check. Each has its own rules, its own log, its own idea of your customer — and none of them answer to one place.
Change means engineering
Want an AI to use a different model, a new play, or a narrower slice of data? On most stacks that's a ticket and a deploy. The people who run the business can't tune the AI that runs the work.
No single record of what it did
When an AI acts, can you see the exact prompt, its reasoning, every tool it called, and the outcome — and replay it? If the answer is scattered across tools, it isn't really auditable.
How an AI is built
Every AI is a blueprint you configure — and a runtime that executes it.
Under the hood, an AI employee isn't a black box you have to take on faith. It's a structured blueprint that spells out, in plain settings, exactly what this AI is:
- Which model — the reasoning engine behind this employee.
- What context — the brief and business facts it starts every job from.
- Which skills — the playbooks it may run — win-back, cross-sell, fraud triage.
- Which tools — the governed, audited actions it may take — never a raw API.
- Which data — the members, segments and metrics it may read — and the personal fields it may not.
Change any of it by editing the blueprint — a new version, with its history attached — and the runtime picks it up. The people who run the business tune the AI that runs the work, without waiting on a deploy.
One action, fully recorded
The whole trajectory, in one entry.
Every step lands in one append-only entry — stamped with a run ID and the exact blueprint version that produced it.
Prompt
the exact brief and facts the AI saw
Reasoning
why it chose this action, in its own words
Tool calls
every governed action it invoked — and each result
Decision
what it decided to do, or to hold
Outcome
the business result that followed
run #a1f9 · blueprint: Loyalty Manager v4 · replayable
The roster
40+ governed AI actors under one roof.
Organized two ways — by the job they do, and by how they run — so you can see your whole AI workforce the way you'd see any team: by function, and by where they sit in the flow of work.
Marketing & Campaign
Marketing Manager
Turns a goal into a ready campaign — the audience, the offer, the channel, the timing — and hands it up for sign-off.
Loyalty & Retention
Loyalty Manager
Works the win-back worklist one high-value at-risk member at a time, choosing a fitting gesture — or a deliberate hold.
Risk & Compliance
Risk Manager
Guards points from fraud and clears the same-household referral and channel-order review queues within the guardrails.
Member Service
Loyalty Concierge
Answers members in the portal, grounded in their real, consented data — never a guess, never someone else's record.
Content
the AI writing team
Drafts on-brand emails, subject lines, push copy and creative from your brand kit and live data — a reviewed starting point.
Customer Intelligence
the behavior & prediction helpers
Read the Consumer World Model — churn risk, purchase timing, predicted value — so every other AI acts on a real signal.
Analytics
the metric helpers
Answer business questions from the governed semantic layer — one certified definition, no invented numbers, no raw SQL.
Research
MarPro & the synthetic panel
Run the research and rehearse a campaign on a digital twin of your members before a single real message goes out.
Organized the second way
The same governed AI, by how it runs.
One roster, two lenses. However an AI runs, it's the same blueprint discipline and the same event log underneath.
As named employees you manage
The four managers and the concierge hold a real job — you give them goals, budgets, tasks and feedback, and manage them like a team. One case at a time, inside the limits you set.
As worklist workers
Behind the queues — win-back, fraud, receipt and order review — an AI claims one case, decides inside the server-side guards, and completes it or escalates to a person.
As embedded helpers
Threaded through the product: the writing team in the campaign canvas, the offer advisor at the coupon, the metric answerer in a dashboard — the same blueprint discipline, in the flow of the work.
Meet the named employees on SoClaw, and see the plays they run in the Enterprise Skill Library.
Skills
Give an AI a skill the way you'd give an employee an SOP.
A skillis a structured playbook: when to act, which steps to take, and the guardrails around them. You compose skills onto an AI's blueprint the way you'd hand a new hire the standard operating procedures for their role — so what runs isn't “whatever the model came up with,” but a named procedure your team can read, version and approve.
The library ships with the retention playbook already written, and every skill declares the member state it needs to run responsibly — failing closed to a person when the signal isn't there.
Composing skills onto the Loyalty Manager
Churn Early Intervention
step in while a small gesture still works
Win-back Worklist Execution
one at-risk member at a time, inside the guards
Coupon Budget-Safe Issuance
never past the live per-store cap
Compliance Outreach Pre-flight
consent & quiet-hours checked before any send
Illustrative composition · each skill is versioned, verifiable and state-gated
One event log
Every action an AI takes lands in one place — and stays there.
Whatever the AI, however it runs, every action writes to a single append-only event log: the prompt it saw, its reasoning, every tool call and its result, the decision it made, and the business outcome that followed. Each entry is stamped with a run ID and the exact blueprint version that produced it.
Because the record is complete and versioned, the same entry is auditable (you can see exactly what happened and why), replayable (you can reconstruct the run), and reviewable(a person can check the AI's work, one entry at a time).
Event log · append-only
Illustrative trajectory · entries are never edited, only appended
The safety floor
Five guardrails that are always on — and never configurable.
You tune what an AI does — its model, context, skills, tools and data. You never tune the safety floor. These five run under every AI, on every action, and no setting can switch them off.
Authorization
Every action carries a per-action gear — automatic, human review or forbidden — and anything that messages a customer is permanently locked to human review.
Spend caps
Coupons and points are issued only against live per-store budgets with hard ceilings — an AI can never spend past the cap, by construction.
Budget
Each AI runs inside a token and cost budget, pre-charged before it acts, so an agent can never run away with your spend.
Kill-switch
Any AI can be paused instantly — off by default, capped when on, and stoppable the moment you want it to stop.
Audit
Every step — prompt, reasoning, tool call, decision, outcome — is written to the append-only event log. Nothing an AI does goes unrecorded.
Tune the behavior, never the floor
The whole point of a governed platform: the surface you configure and the safety underneath are two different layers — and only one of them is yours to change.
The same discipline applies to any AI you connect.
The governed platform isn't only for Flash's own employees. Any AI assistant your team connects runs through the same blueprint of tools and data, the same skills, the same approval gears and the same append-only event log — reading your business through one governed interface, never raw tables, and never the personal fields it isn't allowed to see.
FAQ
The questions teams ask before they hand AI real work.
Do we need engineers to change what an AI does?
No. Each AI is a blueprint — model, context, skills, tools, data — that you configure in settings. Changing what an AI can do is a new blueprint version, not a code change or a deploy.
What exactly can we see after an AI acts?
The whole trajectory in one append-only entry: the prompt it saw, its reasoning, every tool call and result, the decision, and the business outcome — stamped with a run ID and the exact blueprint version, so it's auditable, replayable and reviewable.
Can we tune the safety guardrails to move faster?
No — and that's the point. Authorization, spend caps, budget, kill-switch and audit are always on and never configurable. You tune what the AI does; the safety floor underneath is not a setting anyone can lower.
How is this different from the approvals console?
AI Governance is one guardrail — the per-action authorization and approvals layer. This platform is the whole runtime: how every AI is built as a blueprint, composed from skills, run across surfaces, and recorded in one event log, with governance as one of its always-on floors.
Related reading
Keep exploring the pages most related to this one.
Intelligent Decision
A next-best-action: who, which offer, which channel, when.
Read more CapabilityConsumer World Model
The core technology under every decision: a Behavior Model that learns how your members actually behave — trained per brand on their own event stream, honesty-gated against a transparent baseline — plus a Digital Twin that simulates what they'd do next, rehearsing campaigns (and the option of doing nothing) before a message goes out. Together, one complete, living model of your member base.
Read more CapabilityDashboard
Real-time scan-to-redemption data as KPI trends, funnel diagnostics, and a rule + LLM next-best-action panel — plus retention curves by signup cohort and a 14-day revenue outlook that's honest about being a projection, not a commitment.
Read more CapabilityBusiness Goals
Turn any governed KPI into an operating goal that answers to someone. Break one target down on a single parent-child tree across two axes — time (year → quarter → month) and organization (company → store regions → store) — then let actuals roll up automatically from real orders on the same caliber your dashboard reads, with red / yellow / green status, a pace projection, and an alert the moment a goal turns off track.
Read more CapabilitySoTag for Slack
The Slack agent that hands governed, audited numbers to Claude in-thread: daily anomaly scan + human-approved coupon-send. Early access.
Read more CapabilitySoClaw — Hire your AI employees
Hire AI employees for your loyalty program — autonomous agents that each hold a real job: a Loyalty Manager that wins back at-risk customers, an Order Manager that clears the receipt and order review queue, a Risk Manager that protects points from fraud, and a Marketing Manager that turns a goal into a ready campaign. Each works one case at a time within limits you set and logs every decision; you manage them like a team with goals, budgets, tasks and feedback. Off by default, capped, instantly pausable. Early access.
Read more