TE
Travis Eric
HomeNow
Work with meBook a 30-minute call
Travis Eric LogoTravis Eric

AI consulting and custom software, based in Fort Collins. Explore the work. Bring your next idea.

Travis Eric AI Consulting(720) 770-2979

Navigate

  • Home
  • Now
  • Projects
  • How I work

Deeper

  • Work with me
  • About
  • Share your experience
  • Case Studies
  • Resources
  • Impossible Vision
  • Technical Stack
  • Manifesto
  • Writings

Follow the work

New systems and field notes, sent when they ship.

Communities

AI Builder's LabView membershipTeneo Publishing CollectiveView details
@TravisEric_Travis Eric@TravisericLive Builds@travis_eric
PrivacyTerms© 2026 Travis Eric

Now · Oct 11, 2026

One system for local businesses, everything else serves it.

The story is further downSee Riley reach real use now.json

5 client sites live2 paying clients112 real customer calls answered by Riley24 confirmed bookings35 real leads6 owners using their dashboardmeasured Oct 11, 2026, last 90 days

The machine

How the AI build system actually performs.

From the Goal Autopilot's own ledgers, which begin Jul 10, 2026 when it started recording; exported Oct 11, 2026. Counts only: no task, repository or machine names.

Autonomous attempts

3,330 attempts by AI agents working alone. 526 landed; 1,895 were stopped by their own checks.

Hover or tap a week.

Week of Jul 6Week of Oct 5
  • Landed
  • Salvaged
  • Held for a check
  • Stopped

Each bar is one week of attempts, coloured by how each ended. Stopped means a verifier, a test or a missing dependency refused it before it reached the main branch, which is the system working as designed. Attempts peaked in July (2,159) and ran 254 in October.

Why attempts stopped

The most common stop: the verifier said no.

  • The verifier said no895
  • Needed a repair pass483
  • Unclassified400
  • Waiting on another task299
  • Landed, not yet marked251
  • Needed me138
  • Stalled out131
  • Tests failed78
  • Could not prove where it came from46

Every attempt that did not land, grouped by the reason the system recorded. Only a small share needed me.

Finished, and how far it proved

1,962 tasks officially finished. 1,143 proved they were built; only 106 proved they work end to end. That gap is the job.

Hover or tap a week.

Week of Jun 29Week of Oct 5
  • Built
  • Deployed
  • Works
  • Fed
  • No rung recorded

Each finished task records the highest rung it proved: built, deployed, works (a real call went through and was checked) or fed (real data flows). Dark means the task predates that record.

Who did the work

Codex (GPT) ran the most attempts, averaging 25 minutes each.

  • Codex (GPT)2,035
  • Unrecorded520
  • Claude Sonnet485
  • Claude Opus171
  • Claude Haiku69
  • Claude Fable50

Machine A 55% · Machine B 43% · Machine C 1% · Machine D 1% of 7,232 dispatches

Attempts by model family, and dispatches split across the 4 machines in the fleet. Minutes are wall-clock time per attempt.

One system · five receipts

From a working phone line to real use.

Across 4 client businesses, Riley answered 112 non-test calls in a 90-day snapshot. Separately, WorkforceOS recorded 2 appointments booked by phone.

Read the stages as stronger evidence for the shared Riley system. Blue marks deployment, gold marks a production test, and green marks real use. Dates are observation dates, not live health lights.

Receipt 05 / 05 · real traffic

2 bookings

Client booking rail · observed Oct 11, 2026

WorkforceOS recorded 2 non-test appointments booked by Riley on the phone.

Counted read-only from production Booking records; test and canary appointments were excluded.

This is a separate aggregate from the call count. It is not a caller-level conversion rate or a booking on Travis’s consulting line.

Source · WorkforceOS Booking, read-only production count

Read all five receipts as text
  1. BUILT: Found the exact slot · 9:00 AM

    Riley offered the real open Wednesday 9:00 AM slot requested on a production call. The booking tool read WorkforceOS availability and returned the exact start time after the voice-provider field mapping was repaired.

    The caller did not confirm a booking in this check.

    Sep 13, 2026 · Travis consulting line · Sep 13 production call and calendar readback

  2. DEPLOYED: On the consulting line · Live line

    Riley answered a production call on Travis’s consulting number. The call ledger and live agent readback show the published number routing to Riley’s greeting and booking tool.

    A live line does not mean a real discovery call was booked through it.

    Sep 13, 2026 · Travis consulting line · Sep 13 consulting-line call and agent readback

  3. WORKS: Checked a booking · 2 of 3 checks

    Two rapid production canary calls wrote a booking and then canceled it. The canary crossed the phone-to-WorkforceOS boundary, read back a real booking ID, and cleaned up the appointment.

    One of the three rapid runs failed and paged. These were synthetic calls, not customer bookings or a current uptime reading.

    Sep 15, 2026 · Travis consulting line · Sep 15 production booking canary receipt

  4. FED: People called · 112 calls

    Riley answered 112 non-test inbound calls across 4 local businesses in the 90-day snapshot. Counted read-only from production voice-call records for ice cream shop, fitness studio, landscaping company, dent and hail repair. Test, QA, canary, spam and outbound calls, our own lines, and test-looking numbers were excluded.

    The count includes five calls from a client owner and some short hang-ups. Calls are not unique customers or qualified leads.

    Oct 11, 2026 · 90-day snapshot · client lines · Conversos voice_calls, read-only production count

  5. OUTCOME: Booked by phone · 2 bookings

    WorkforceOS recorded 2 non-test appointments booked by Riley on the phone. Counted read-only from production Booking records; test and canary appointments were excluded.

    This is a separate aggregate from the call count. It is not a caller-level conversion rate or a booking on Travis’s consulting line.

    Oct 11, 2026 · Client booking rail · WorkforceOS Booking, read-only production count

Aggregate snapshot measured Oct 11, 2026. The checks on Travis's line and the counts on client lines describe the shared engine across different installations; they do not trace one caller. Private call and booking records are summarized here without names.

Did it work

What runs, and what shipped.

Every connection between my systems, how far each has proved itself, and the last 30 days of dated events.

Outcomes for clients, measured

5 client sites live, Riley answered 112 real customer calls for 4 of those businesses, and customers made 24 confirmed bookings in the last 90 days.

Real customer calls Riley answered

  • Ice cream shop45from 19 callers
  • Fitness studio34from 28 callers
  • Landscaping company30from 12 callers

    plus 5 from the owner, not counted

  • Dent and hail repair3from 3 callers

Google reviews since tracking began

  • Bar and grill+361,184 → 1,220, Jul 4–Oct 5, 4★
  • Ice cream shop+14333 → 347, Jul 11–Oct 5, 4.9★
  • Burger restaurant+102,451 → 2,461, Sep 12–Oct 5, 4.6★
  • Dent and hail repair+4134 → 138, Sep 3–Oct 5, 5★
  • Chiropractor+027 → 27, Sep 12–Oct 5, 5★
  • Title company+026 → 26, Sep 12–Oct 5, 4.7★

Counts were first recorded when tracking started, not when the work began, so these are gains since then.

on the client’s own domain
client sites live
5
each with a payment receipt
paying clients
2
2 booked by Riley on the phone
confirmed bookings
24
form and booking requests, 5 businesses
real leads
35
on 2+ days in the last 30
owners using their dashboard
6
since tracking began, 6 businesses
Google reviews gained
+64

Paying clients, and how each payment was verified: dent and hail repair since Sep 8 (Venmo, recorded with its invoice); landscaping company since Sep 13 (Stripe subscription, live, invoice paid). Amounts are not published.

Measured read-only from the production call, lead, booking, dashboard-activity and review records on 2026-10-11, by a script that runs daily. Excluded: anything flagged as a test, probe or canary, robocalls, outbound calls, our own phone lines, my own numbers, owners calling their own line, and test-looking names and emails. Businesses are described by trade, not named.

How far each connection has proved itself

Of 77 connections between my systems, 63 have been deployed and 53 pass an end-to-end check, but only 19 have been proven with real traffic. Most of the gap is test traffic waiting for real users.

  • Tracked connectionsevery system-to-system link on the board
    77
  • Builtthe code exists and passes its own checks
    76
  • Deployedrunning in production
    63
  • Works, checked end to enda real call went through and was read back
    53
  • Proven with real traffica real customer, owner or visitor, not a test
    19
  • Fed by real datareal data flows through it routinely
    12

Connections needing attention or an intentional restart

  • Conversos → Client Factory · blocked: The feedback receipt worked on one consented call. The full journey stops before a verified owner confirms the brief.Next: Run the supervised verified-owner journey and read the confirmed brief and owner interface back.
  • Ice Cream Magazine → Teneo Publishing · blocked: The last Teneo fallback probe authenticated but reported insufficient upstream credit. The current primary editorial desk uses signed-in Codex and Claude subscriptions, so this is not a magazine-wide outage.Next: If the paid fallback is needed, restore provider credit and probe that route specifically; do not use the primary subscription probe as proof of this fallback.
  • Orchestrator → Error Scope · paused: Cron monitoring still runs in Supabase; durable ClickHouse exception and check-in history was intentionally turned off for cost.Next: Keep cron control-plane checks; restore ClickHouse ingestion only after the cost decision and a named owner.
  • Orchestrator → LLM Scope · paused: Trace storage was intentionally turned off with ClickHouse; producers fail open.Next: Restore storage only after the cost decision and verify a new span from the receiver.
  • Roomly → Error Scope · paused: The client queues reports while ClickHouse ingestion is intentionally off.Next: After the cost decision, restore ingestion and read a queued report from Error Scope.
  • teneo-auth → Client sites · split: The generic send route uses Resend while owner invites and magic links can use SES. This is a routing inconsistency, not evidence of a delivery outage.Next: Move approved generic sends to the SES rail named in the communications SSOT, then read a message back from an owned mailbox before cutting over each sender.

Each bar counts the connections that reached at least that level, including a connection whose current condition has since changed. Works means a check crossed the connection and read the result back; test traffic (canaries, flagged probes, my own calls) is counted separately from real use. 3 intentionally paused, 2 blocked by prerequisites, 1 with split routing. Rungs reviewed 2026-09-25; these conditions were reviewed 2026-09-27. Source: the maintained ecosystem integration registry and the ClickHouse pause receipt.

The system map

31 systems, 56 links between them. Client Factory is the most connected, with 19.

Tap a system to light up its connections; tap it again to clear.

Client FactoryTeneoConversosAdFoundryClient sitesaibridgeMarketingOSAnthropic APIAnalyticsOSaudio-engineFortListOSDetection Labteneo-authFinForensicsWorkforceOSFort List appAI assistantKnowledgeOSError ScopeLLM ScopeOrchestratorPrintfulLife ClassProfileEnginetraviseric.comRoomlyTrendOSTeneo Publishingbrowser-engineTeneo VoiceIce Cream Magazine
  • Fed by real data12
  • Works with real traffic7
  • Works on test traffic34
  • Deployed, not yet proven10
  • Built, not deployed13
  • Planned1
  • Paused, blocked or split6

Each dot is a system, sized by how many connections it has. Each line is a connection, coloured by the furthest it has proved itself: green is fed by real data, solid gold works with real traffic, dashed gold works on test traffic, blue is deployed, grey is built. A red ring marks a current limitation; inspect the connection for its specific condition.

TrendOS, reading the world

711,853 signals collected from 64 sources.

711,853

signals on record

  • Fresh in 24 hours: 42
  • Fresh in 7 days: 1
  • Older: 21

Latest signal landed Oct 11. Collection is running.

TrendOS is my research engine: it collects signals from public sources and scores them. The bar shows how recently each source last delivered, as of TrendOS's last stats refresh.

The last 30 days

5 rung climbs, 7 launches and 21 field notes in 30 days.

Rung climbs
Launches
Writing
Sep 12Sep 27Today

Each mark is one dated event, oldest on the left and today on the right. Hover a mark to see what it was. The launches and notes are listed further down.

Beyond GitHub

Every product, start to now.

From the AnalyticsOS registry, refreshed hourly.

Product lifelines

16 products in the registry: 10 launched, 6 still in development, 10 touched in the last two weeks.

2025
2026
  • Teneo
  • ProfileEngine
  • Conversos
  • AnalyticsOS
  • Image Engine
  • MarketingOS
  • Orchestrator
  • Teneo Auth
  • Browser Engine / Detection Lab
  • Teneo Publishing
  • FitLabs
  • DomainOS
  • FormForge
  • Revenue Engine
  • OpenBazaar AI
  • AI Bridge

Each row is one product from the AnalyticsOS registry. The thin line runs from its start to launch, the thick bar from launch to its latest commit; blue rows are still in development. Start and launch months are as registered.

The long version

What I'm focused on.

Most of my attention now goes to one thing: the system that runs a local business’s website, phone, leads and bookings, sold through Local Business Factory. Client sites are live on it, Riley answers their phones, and the first clients pay for it. The numbers are measured on this page every day.

The offer is now three fixed plans instead of custom quotes, and the next job is the ads and funnel that bring owners to them. My other tools, the book engine included, now earn their place by serving those clients rather than as products of their own.

I decide what to build, set the checks each change must pass, and review what lands. AI coding agents write most of the code, which is why the commit counts run high. The outcome numbers, sites live, calls answered, bookings and owners using their dashboards, are the ones that matter.

Focus note reviewed Sep 28, 2026.

Progress

Measured in what actually runs.

Commit counts say I was busy. This board says whether anything works. A rung only counts when the check behind it has been run.

BUILT · 2
The code exists and passes its own checks.
DEPLOYED · 0
It is running in production.
WORKS · 5
A real call went through it and the result was read back.
FED · 1
Real, non-synthetic data has flowed through it.
  • Operator Guild

    Another owner's AI visits mine and asks it questions, each side running on its owner's own machine.

    Open the door
    1. BUILT
    2. DEPLOYED
    3. WORKS
    4. FED

    synthetic visitors · since Sep 23

    The front door, mailbox and host all answer, and a test visit got a real reply back.

    Next rung · The first real visitor. One invite is out.

  • Riley on my consulting line

    An AI receptionist that answers my business number, explains the work and books calls.

    Call (720) 770-2979
    1. BUILT
    2. DEPLOYED
    3. WORKS
    4. FED

    since Sep 13

    Asked for "Wednesday at 9", she found the real open slot on my calendar and offered to lock it in.

    Next rung · A caller booking a real discovery call through her, start to finish.

  • TrendOS research feed

    Finds and qualifies local businesses and feeds them to the prospecting system.

    Open TrendOS
    1. BUILT
    2. DEPLOYED
    3. WORKS
    4. FED

    research reads · since Sep 8

    Real business records from two scheduled cloud runs landed in the prospect roster.

  • The Fort List newsletter

    Weekly Fort Collins events email with its own consent, welcome and unsubscribe path.

    Open The Fort List
    1. BUILT
    2. DEPLOYED
    3. WORKS
    4. FED

    since Sep 12

    Signed unsubscribe and scheduled sends proven end to end with test readers, then cleaned up.

    Next rung · An edition reaching a real subscriber. The first send had nobody eligible yet.

  • Fort List and the ice cream magazine, talking

    Two separate products share events and guides without sharing a database.

    1. BUILT
    2. DEPLOYED
    3. WORKS
    4. FED

    since Sep 22

    Verified in both directions on my machine.

    Next rung · Deploying both sides.

  • The Fort List iPhone app

    Events, sign-in, interests and saved lists, on the phone.

    1. BUILT
    2. DEPLOYED
    3. WORKS
    4. FED

    Every screen works against the live production service.

    Next rung · App Store approval. Review asked for more information about the app first.

  • My own booking calendar

    The "book a call" rail on my sites. I built it instead of renting Calendly.

    See the calendar
    1. BUILT
    2. DEPLOYED
    3. WORKS
    4. FED

    since Jul 24

    Bookings write the calendar slot, the contact record and the analytics event.

    Next rung · A stranger booking through it, counted.

  • Consulting enquiries into my CRM

    A written enquiry on this site becomes a lead record, not an email in a pile.

    Open the enquiry form
    1. BUILT
    2. DEPLOYED
    3. WORKS
    4. FED

    since Jul 27

    A flagged test enquiry was read back from the CRM and deleted.

    Next rung · A real enquiry, read back the same way.

“Fed” means real, non-test data has flowed through it. It never means revenue, and nothing on this page claims a client result. For openable demos, see the proof page.

Shipped · last 30 days

Things you can open.

  1. Sep 28

    Client outcomes, measured

    Sites live, real customer calls, bookings, leads and paying clients, counted daily from production records instead of typed by hand.

    See the outcomes
  2. Sep 23

    The Operator Guild opened

    An invite link that lets another owner's AI visit mine. The page explains itself to any assistant that lands on it.

    Open the Guild
  3. Sep 22

    Every /proof link re-checked

    All 11 live systems on the proof page were re-requested and answered, so the page carries one verified date.

    Inspect the proof
  4. Sep 16

    The marketing system, piece by piece

    Every part of the marketing system on one page, each with the honest rung it has actually reached.

    Read the system
  5. Sep 15

    Judgment, the argument

    Why judgment is the scarce thing now that output is cheap: five skills, each with the date I learned it the hard way.

    Read the argument
  6. Sep 15

    Writings rebuilt as field notes

    Moved off the old CMS onto plain files. Only posts I wrote or approved are public; the rest stay behind a gate.

    Read the notes
  7. Sep 13

    Riley books the real open slot

    Calls were going unbooked because of a field-name mismatch with the voice provider. Fixed, and she now offers the exact open time you asked about.

    Call her

Written · 21 field notes

  1. Oct 11Dispatcher
  2. Oct 10Copy Arena — draft, read, refine, launch
  3. Oct 9signup-doors-walk: every door, including the moment after submit
  4. Oct 8goal-finder — derive the goals a repo needs to finish
  5. Oct 7Motion Studio: the video harness
  6. Oct 6section-pass — one agent per section, the details put in
All writings

Open loops

What is not done yet.

  • Making the dashboard earn its keep for every owner

    Two owners now use their dashboard on most days. Several others found it confusing at first, so the work is making it clear enough to use unaided before I ask any client for a referral.

  • The Fort List app is waiting on Apple

    Review came back asking for more information rather than a fix. The reply and a device recording are next, then resubmission.

  • The Guild needs its first real visitor

    Everything answers with test traffic. It proves nothing until another owner actually sends their AI through.

Not doing

What I chose to leave alone.

  • Selling Teneo on its own

    There are a hundred book generators now. Its tools stay, but they earn their keep by serving local-business clients rather than as a separate product.

  • A content treadmill

    No video schedule. I write when something was learned the hard way, and only then.

  • Renting the booking widget

    Every "book a call" runs on the calendar I own, so the clients own their data too.

Work with me

Three ways in. Or call the line and let Riley explain it first.

  • The Clarity Pass

    One week, fixed scope.

    I read your page or pitch cold, rewrite it, audit every claim on it, and hand back the receipts.

    See the week
  • Custom AI and software work

    Scoped on a 30-minute call.

    Larger or less defined work: what to build, whether it is worth building, and then building it.

    Book a call
  • Local business plans

    Monthly plans for local businesses.

    The system on this page, run for your business through Local Business Factory: site, phone, leads and bookings.

    See the plans