An AI automation consultant in San Diego should split pipes from agents, then write one leftover into tools you already pay for. Nidrosoft is the San Diego studio of Cyriac Ndo Zeh, a design engineer who implements AI for small and mid-size businesses: audits, custom builds, phone and receptionist agents, training, and systems the team can own.
Written by Cyriac Ndo Zeh, Founder of Nidrosoft — 12+ years building software and AI products for San Diego small businesses. Also on LinkedIn.
Key takeaways
- Process automation moves a known object on a known path. An AI agent handles messy language, then writes a structured row a person still gates.
- Industries this studio already shapes work for: dental, property, dealers, software teams, contractors, and desks with a published phone.
- Before and after examples below are labeled composites, not named clients. Public systems: Omnira Dental, Eviction Wizard, TokenTra, Nexuvo.
- Typical bands: readiness call free; audit often low thousands; first custom agent often mid four- to five-figure. A $50/mo receptionist SaaS plan is a different leftover.
- Fit is usually 30-250 people, plus smaller shops if the workflow is clear.
- Book a free 15-minute AI readiness call before you buy another Zap that still needs a human paste.
Contents
- Process automation versus AI agents
- What an AI automation consultant changes in a week
- Industries and desks we actually serve
- Before and after workflows
- DIY tools versus a SaaS chatbot versus a custom agent
- Where Zapier ends and a write path starts
- San Diego texture on the automation row
- How the engagement runs
- Frequently asked questions
- Name the path, then book
Process automation versus AI agents
Process automation is a pipe. When EstimateStatus becomes Sent, text the owner. When a form hits Gmail, create a HubSpot lead. When a token invoice lands, tag the feature in a sheet. The object is known. The path is known. Zapier, Make, native HubSpot workflows, and ServiceTitan automations already do this work. Buy them when the field already exists and the rule is a sentence.
An AI agent is a reader and a writer on messy input. A caller describes a no-cool in Spanish. A classified listing is a paragraph with a price buried in the third line. A patient replies “after the kids’ soccer, maybe Thursday.” The agent turns that language into JobType, AskPrice, RecallDueOn, LanguagePref. A person still approves money, filings, and clinical exceptions.
Mixing the two words on one invoice is how you pay for a model to move a field that a checkbox already moves. It is also how you buy a Zap and expect it to understand a voicemail.
The consultant’s first job is the split. Hours on deterministic work go to the pipe. Hours on language and judgment go to the agent, with a human gate. Hours on work that happens once a year go back to a person.
A homepage chatbot is an agent only when hours and access support it. Most unpaid work in this county is recall, follow-up, after-hours capture, inventory hunt, notices, or token spend. Those leftovers need a system of record, not a bubble.
The method under the split is Embed, Diagnose, Scope, Build, Own. The pillar sequence is AI implementation for small business. Stay here when the search term is automation and the constraint is “we already have Zaps and Friday still closes by hand.”
What an AI automation consultant changes in a week
Week one is Embed. Sit with the people who already run the unofficial path. The group text. The spreadsheet. The ChatGPT paste. The Copilot draft that never became a row. Map official automations you already pay for so the later write does not duplicate them.
A useful week one artifact is a two-column list. Column A: pipes that already work. Column B: language jobs that still eat hours. If leadership only allows a tour of the official SOP, the week is unfinished. We will say so rather than invent hours.
Week two is Diagnose. Each item in column B gets hours, feasibility, and payback. Feasibility includes API reality and whether counsel will allow customer text in a third-party model. Payback includes time returned and errors avoided. Favorite chatbot ideas go last unless they win those three.
You leave with field names and a recommended next purchase. That purchase might be a better Zap. It might be seats and a workshop. It might be a custom agent. It might be stop.
Cyriac is the person in the sitting. 12+ years, 125 products shipped, 16 public. Formerly Anthropic, Microsoft on Bing and Edge, Intuit, and Gap Inc. There is no junior bench that “handles automation” while a partner flies in.
Typical bands: readiness call free; scoped audit often low thousands; first custom agent often mid four- to five-figure depending on systems. This page is not a quote.
For a city-shaped automation map, read AI automation San Diego. Use that page when the search term is the county. Stay here when the search term is the consultant leftover.
Industries and desks we actually serve
This studio fits companies of about 30-250 people, plus smaller shops if one workflow is clear. The industries below are desks we can name without inventing a client list.
Dental and clinic fronts. Omnira Dental is the public practice-shaped system. Hygiene recall, intake, confirmation. A coordinator still sends. A clinician still charts. Open Dental, Dentrix, EagleSoft.
Property desks. Eviction Wizard is the public packet-shaped system. Notice-to-writ assembly. A person still files. AppFolio, Buildium, a shared inbox.
Car dealers. Nexuvo is the published hunt: 3x inventory, 80% less manual outreach. A person still buys the car. Dealer CRM, classified sites.
Software teams. TokenTra shows spend by feature. Teams using that class of visibility have cut 10-30% of wasted AI spend. Bring invoices.
Contractors and home services. Follow-up and after-hours capture. ServiceTitan, Jobber, Housecall Pro, a zoned board. The HVAC vertical is a sibling leftover. Law firm intake is another: read AI agent for law firm when the language job is a matter screen, not a truck.
Hospitality and retail desks appear when the leftover is reservations, no-shows, or a published number. Gaslamp nights are a clock, not a vibe.
We do not publish a hospital-wide automation program. Past 250, pick one department. We do not publish a once-a-year board process as automation.
Workshops have been run in La Mesa. Training is a leftover when the team needs judgment on which jobs stay pipes and which jobs need an agent.
Before and after workflows
The stories below are labeled composites of typical shops. They are not named customers and not testimonials.
Typical La Mesa dental desk, before. The hygiene coordinator exports a recall list on Monday, pastes names into ChatGPT for reminder drafts, copies each draft into the PMS message tool, and marks a paper list. After. A queue shows RecallDueOn, LastAttemptAt, AttemptChannel, DoNotContact. Drafts land next to the row. The coordinator still sends. The clinician still owns the chart. Omnira Dental is the public shape for this leftover.
Typical Kearny Mesa lot, before. An inventory buyer opens Facebook Marketplace, Craigslist, and Cars.com after dinner, messages sellers by hand, and pastes survivors into a notes app. After. A hunt writes SourceUrl, AskPrice, Mileage, NeedsHumanBuy, CrmRowId. Outreach volume drops. A person still buys the car. Nexuvo is the published pattern (3x inventory, 80% less outreach).
Typical East County property desk, before. A coordinator rebuilds notice packets from email threads and a spreadsheet of ServedOn dates. After. A packet path assembles UnitId, NoticeType, ServedOn, ReadyToFile, CounselSignoff. A person still files. Eviction Wizard is the public shape.
Typical UTC software team, before. Model invoices arrive as a single number. Nobody can say which feature wasted the spend. After. TokenTra-shaped visibility tags Vendor, FeatureName, TokenCount, WastedFlag, BudgetOwner. A person still sets the budget. Published cut on wasted spend in that class of tool: 10-30%.
Typical Oceanside HVAC shop, before. After-hours calls dump to a cell. Quotes and no-cools look the same at 11 p.m. After. An agent collects JobType, Zip, UrgencyFlag, LanguagePref, AcceptNextMorning. Phrase list pages on-call. Filter requests wait for 7 a.m. A person still accepts the emergency truck.
Typical Chula Vista front desk, before. An English-only prompt loses Spanish-first callers. After. LanguagePref and ConfirmLanguage stay one language through the hold. A person still takes named-person requests.
NanoBrief is the briefing composite: under 2 minutes versus 3+ hours. Use that speed for a scoped packet. Do not buy a monthly chair for an annual job.
DIY tools versus a SaaS chatbot versus a custom agent
Automation buyers already own seats and Zaps. The comparison table still earns a place because the next invoice is usually one of three rows.
DIY ChatGPT or Copilot: drafting speed. Same day. Prompts you save. Vendor help center.
Generic SaaS chatbot or receptionist: pickup. Same week. Script templates. Vendor ticket. Some plans near $50/mo.
Custom Nidrosoft agent: language in, trusted row out. Audit then write. Pairing and files you own.
| Approach | Typical cost | Setup time | Customization | Support |
|---|---|---|---|---|
| DIY ChatGPT or Copilot | Seat licenses | Same day | Prompts you save | Vendor help center |
| Generic SaaS chatbot or receptionist | Low monthly leftover; some market near $50/mo | Same week | Script templates | Vendor ticket |
| Native Zaps and workflows | Existing stack | Days | Known fields, known paths | Vendor plus your admin |
| Custom Nidrosoft agent | Audit in the low thousands; first agent mid four- to five-figure | Two-week audit, then a written build | Messy language to a trusted row | Pairing, runbook, files you own |
Buy the Zap row when the field already exists. Buy the custom row when the input is a paragraph, a voicemail, or a classified listing. Buy the seat row when a human will still paste. Buy the SaaS row when pickup this week is the leftover.
Where Zapier ends and a write path starts
Zapier ends when the trigger is clean and the action is a known field. A form submit to HubSpot. A paid invoice to a Slack channel. A calendar cancel to a no-show tag.
A write path starts when the trigger is language. “The AC died and my mom has a breathing machine” is not a dropdown. “Is the 2014 Camry still available, cash today, 142k miles” is not a form. “Don’t text me, we moved to Oceanside on orders” is a DoNotContact plus a Transferred flag, not a marketing nurture.
The consultant who automates the second set with a Zap and a hopeful prompt will ship errors into production. The consultant who puts a model on the first set will bill you to move a checkbox.
Eval the language paths. Spanish confirmation. After-hours phrase list. Money and filing gates. DoNotContact. If the model cannot fail closed, it does not write.
Keys in your vault. Prompts next to the code. Nerlude exists because founders lose credentials in chat.
San Diego texture on the automation row
North County confirmation texts. South Bay Spanish-first callers. East County truck follow-up. Kearny Mesa night inventory. Gaslamp no-shows. Carlsbad and Oceanside tourism overflow. La Jolla and UTC inboxes next to biotech. Military families who break lists. Interstate 8 and I-15 clocks on when a 4:30 yard close happens.
An automation consultant who sells the same first Zap to Encinitas, Chula Vista, and El Cajon without watching Friday is selling a template. Templates belong to the SaaS row.
Commute time is part of the calendar. Remote audits are normal. In-room days help when the unofficial stack is a clipboard. Workshops have been run in La Mesa.
We will not take a brief that forbids the unofficial path. We will not take five agents in month one. We will not promise full autonomy on filings or buys.
How the engagement runs
Readiness call, free. Assessment, often low thousands. Written go or stop. Build in your tenant. Own with a revoke test. Scale only after the thirty-day metric, with Diagnose reopened in writing.
Public systems you can open before you book: Omnira Dental, Eviction Wizard, TokenTra, Nexuvo, Octana, Guidera, Protectron, Nerlude.
Calendar: https://calendar.app.google/GCtfuTx3Ms6K7d579. Email: cyriac@nidrosoft.com.
Eval the language paths before you call the leftover done. Spanish confirmation stays one language. After-hours phrase list pages only on written phrases. Money and filing gates fail closed. DoNotContact and Transferred stop the sequence. A model change that books Sunday on a closed board fails the eval.
A typical Encinitas clinic composite, labeled as typical, not a named client. Before: the coordinator pastes recall names into ChatGPT, then into Open Dental, then onto a paper list. After: a queue with RecallDueOn and LastAttemptAt. The coordinator still sends. That is an agent leftover sitting on top of a PMS, not a Zap that already had a clean trigger.
A typical El Cajon contractor composite. Before: estimate follow-up lives in a truck group text. After: LastTouchAt and EstimateStatus on a row the office can sort on Monday. If Jobber already moves “Won” to a tag, keep that Zap. The agent belongs on the reply that says “maybe Thursday after the inspection.”
Bring invoices if spend is part of the automation story. TokenTra-shaped visibility is a pipe once features are tagged. The language job is the Slack dump that never became a row.
Frequently asked questions
Do I need an AI automation consultant if we already pay for Zapier?
Yes, if Friday still includes language jobs: voicemails, classified listings, patient replies, Spanish callers. No, if every unpaid hour is a known field on a known path. Assessment will say which column you are in. Keep the Zaps that already work.
What is the difference between an agent and a chatbot?
A chatbot answers. An agent writes a structured row into the system of record and stops at the human gate. If nothing is written that a coordinator trusts, you bought a conversation. Homepage widgets lose this rank unless hours and access support them.
How much does this cost?
Readiness call is free. A scoped audit often lands in the low thousands. A first custom agent often lands in a mid four- to five-figure project band depending on systems. A $50/mo receptionist SaaS plan is a different leftover. This page is not a quote.
Which industries do you automate?
Desks we can name: dental and clinic fronts, property, dealers, software teams, contractors and home services, law-firm intake when the leftover is a screen, hospitality when the leftover is the book. Fit is 30-250 people, or smaller if the workflow is clear. Past 250, pick one department.
Will automation replace the coordinator?
The coordinator still sends, files, buys, and handles exceptions. Automation removes the rebuild of the same list and the paste into the same tool. If your goal is a headcount cut, say that in the readiness call. The audit may refuse a scope that hides the human gate on money or clinical notes.
Name the path, then book
If the object is known, buy a Zap. If the input is language, book the call. If you cannot tell, book the call and we will split the columns.
Book a free 15-minute AI readiness call. Write cyriac@nidrosoft.com if the calendar does not fit. Bring the unofficial path, the automations you already pay for, and the owner. The leftover is a pipe or an agent. Write that on the invoice before anyone maps a field.