Deep Dive
How your workforce
actually works.
The architecture is real. There are policies you set. There are runbooks the agents follow. There is a daily cycle. There is a self-improvement loop.
The agent daily cycle
Every agent runs a continuous lifecycle, not a script that fires when you click a button. They wake up at the start of the business day, plan the day's work against your priorities, execute it across calls, emails, and tools, reflect on outcomes, and improve themselves.
- 1. Wake
- 2. Plan
- 3. Work
- 4. Reflect
- 5. Improve
- 6. Sleep
Autonomy levels: how the two modes work
Every action type gets one of four levels, set per function. Execute and Execute & Notify put a function in run-it-for-you mode; Draft for Review and Escalate put it in run-it-with-you mode. The default for routine work is Execute — the design choice that makes Sentica a workforce instead of an assistant. Promote any function from with-you to for-you with a single setting as trust builds.
Execute
Agent acts immediately within policy and logs silently.
Execute & Notify
Agent acts and adds a line to the morning briefing.
Draft for Review
Agent drafts, queues for owner approval, does not act.
Escalate
Agent pauses and notifies owner; never acts alone.
The Accountability Layer
Anyone can claim autonomy. We stand behind it.
Service businesses are licensed, bonded, and insured. The AI that works for them should be held to the same standard. Five components ship as core infrastructure — and every deployment is backed by a performance guarantee.
Evidence-grade action log
Every action is recorded append-only and tamper-evident, stamped with the agent, the policy version, and the model version. Exportable in formats an accountant, an insurer, or a court will accept.
Deterministic policy engine
Dollar limits, action limits, and prohibited actions are enforced in code outside the model. No clever prompt can talk an agent past its limits.
Reliability ledger
A weekly evaluation scores every agent's actions against real outcomes, producing per-agent trust scores you can see at any time.
Incident workflow
Detection, a kill switch, reversal through compensating transactions (void and reissue, never silent edits), owner notification, and a patch to the responsible skill.
Compliance packs per vertical
Maintained and warranted rule sets your agents must follow, starting with TCPA for calling and texting — where damages run $500 to $1,500 per improper contact.
The difference is accountability, not autonomy. You're never asked for faith — you're shown a record.
The self-improvement loop
Your workforce gets better every week.
Every agent reflects daily, and a weekly evaluation scores its actions against real outcomes. Improvements above a 90% confidence threshold apply automatically. Lower-confidence ones queue for your review. Here are the median results from our design partners across the first 90 days.
Median across design partners, weeks 1–13. Individual results vary by business and runbook configuration.
How it works
Wednesday, 7:14 AM.
Prepares the morning briefing for the owner, Sarah. Three jobs scheduled today, $14,200 in receivables pending, one tight cash week ahead in 9 days.
Opens the briefing on her phone. Two-tap approval queue. She approves a $3,200 vendor payment, taps no on a discount request, and gets back to her coffee.
Answers the first call. New customer, no heat in master bedroom. Books a 10 AM slot with Dave.
Responds to a 5-star review on Google. Personalized. Mentions the spring tune-up promo.
Provisions M365 for the new technician starting Monday. Sends him his welcome kit and login.
Detects the word 'flood' in a call. Pages Dave's backup, Mike, since Dave is mid-job. Quotes the emergency surcharge.
Schedules three interviews for the open service tech position.
Sends invoice $1,420 to the morning customer. Stripe captures payment in 3 minutes.
Publishes the Tuesday blog post: 'When to repair vs replace your AC unit.' Schedules it across social.
Flags Tuesday next week as a tight cash day. Recommends delaying one non-critical vendor payment. Sarah approves with one tap.
Takes an after-hours emergency. Routes it to the on-call tech per the runbook Sarah set up during onboarding.
Sends the end-of-day campaign to 480 customers. 18% open rate, 4 new bookings on the calendar by morning.
All seven agents file their reflections. The self-improvement loop notes Ava handled 14 calls today with zero escalations. Patches her runbook to make a similar response 0.4 seconds faster next time.
That was one Wednesday. Multiply by 365.
See what your Wednesday could look likeThe Architecture
Your AI workforce is not a chatbot. It's a system.
Three things make Sentica work. They are the difference between AI that drafts and AI that does.
You set the rules. The agents stay inside them.
A policy is a guardrail: 'Discounts up to 15% are allowed. Max quote $5,000.' A runbook is a procedure: 'When a customer says flood, page the on-call tech, instruct them to shut the valve, and apply the emergency surcharge.' Together they let the agents operate without supervision. You change a policy in the AI Control Center; the next call respects it.
Your agents wake up. Plan the day. Do the work. Reflect. Sleep. Repeat. Seven days a week.
Every agent runs a continuous lifecycle, not a script that fires when you click a button. They wake up at the start of the business day, plan the day's work against your priorities, execute it across calls, emails, and tools, reflect on outcomes, and improve themselves. This is what makes them feel like employees, not tools.
Your agents work for you. Only you. Forever.
Every customer has their own fully isolated AI agents. Their own identity. Their own memory. Their own runbooks. Their own custom tools. Your Lisa is not the same Lisa another business uses. She has spent every interaction learning your customers, your services, your pricing, your voice. She works for you alone, and no other tenant can ever see her, query her, or touch her data.
Works with whatever you already use.
Tier 1 | Pre-built integrations
For Dialpad, Microsoft 365, Google Workspace, QuickBooks, Stripe, Shopify, and dozens more, we ship the integration. The agent uses it out of the box.
Tier 2 | Agent-built integrations
When you bring a tool we haven't pre-built, the agent finds the API, reads the docs, writes the integration, and gets to work. The integration lives in your tenant and belongs to you.
Tier 3 | Browser automation
When a tool only has a web interface, the agent uses the browser the same way a human employee would. There's no tool the agent can't use.
Your phone system, your choice
Dialpad by default. Microsoft Teams Phone and Zoom Phone natively. Or bring your own — RingCentral, OpenPhone, 8x8, and any modern VoIP. Your provider owns the phone; we own the workforce that uses it.
Your credentials stay yours
Agents retrieve logins at the moment of use from a password vault you own — Bitwarden or LastPass — with per-agent, revocable access. Your passwords never live on our platform.
Your tools, built by your AI workforce.
We don't ship a one-size-fits-all CRM. During onboarding, Lisa builds a CRM tailored to your business: the fields you use, the workflow you follow, the integrations you need. When you say "I want to track which technician serviced each customer last," Lisa extends it. No support ticket, no roadmap wait. The tools live in your tenant. They're yours.