Eric Irish

Eric IrishAustin, TXSeventeen years. No snake oil.

Yes, AI can do real work in your business. I’m the guy who actually builds it.

I embed with how you already work — not to sell you a prettier homepage. Snake-oil salesmen want buzzwords and chat bubbles; I build invisible AI that puts the answer in front of the right person before anyone has to ask. Seventeen years in startups and the systems behind them. I’ll tell you plainly where that earns its keep — and where it doesn’t.

Work with me
Eric Irish in a cowboy hat, suede jacket, and a Texas belt buckle
That’s me. Not an AI photo.
Years building this stuff
17
People you’ll deal with
1
Strategy decks before code
0

01 — Snake oil vs. the real work

What the snake-oil salesman wants

  • Months of “discovery” before anything changes
  • To “optimize you for ChatGPT”
  • Software that only works if someone asks it a question
  • A monthly report full of charts
  • A year-long contract before anything gets built

What actually helps your business

  • Someone embedded long enough to learn how the week actually runs
  • The facts your people keep in their heads — written where the system can use them
  • The stuff you do by hand every day — quotes, bookings, follow-ups — turned into a flow
  • Invisible AI on that flow: the answer already on screen, nobody typing a prompt
  • One person you can call who will tell you no

AI can’t invent a process you never had. If they can’t name the step in your week they’d change — the handoff, the check, the quote — they’re selling snake oil. SEO people have been doing it for years. Now they just say AI.

02 — The thesis

Good AI is invisible. You shouldn’t have to chat with it.

The snake-oil move is making your people ask for information they should already have. “Talk to your business.” That’s a prompt where a screen should have been enough.

What they sell

  • A chat window so someone can “ask the system”
  • A box where your staff has to prompt a robot for last week’s numbers
  • A wrapper on your data that still needs a conversation
  • A demo that looks smart until someone asks a real question

When it’s done right

  • The quote is already drafted when they sit down
  • The booking is already on the calendar
  • The number they needed is already on the screen
  • Nobody typed a prompt. The work just showed up.

You don’t need a chatbot. You need the right thing in front of the right person at the right time — already there, no prompt. Then they get back to the customer.

03 — Ways to work with me

Pick the one that sounds like your week.

  • 01

    One job off your plate

    Quotes, bookings, follow-ups, the questions people call you about. I turn that one job into a system. AI only if it earns its keep — not because a snake-oil salesman said you needed it.

    Good if you have
    One thing on your week that’s eating you.
    Ends with
    That job runs. You don’t hover.

  • 02

    Embed in how you operate

    I work inside your process — quoting, booking, dispatch, compliance rounds, whatever actually runs the week — and wire it so information shows up where people already are. Not a side project. Not a new tool they have to remember to open.

    Good if you have
    Smart people repeating the same steps every morning.
    Ends with
    The flow runs. The answer was already there.

  • 03

    Invisible AI on real work

    The inbox, the spreadsheet, the handoff between teams — that’s where the mess is. I plug AI into those moments so your people get what they need without interviewing a chatbot.

    Good if you have
    You know what should be automatic. Nobody’s had time to build it.
    Ends with
    Information delivered. No prompt.

  • 04

    Keep me around

    After the build, I stick around part-time. When the next snake-oil salesman calls — they usually say SEO or AI — I’ll tell you if they’re selling work or a bottle. When something should be automated, I build it. When it shouldn’t, I say so.

    Good if you have
    After a big operational push, or a team with nobody technical.
    Ends with
    It doesn’t. That’s the point.

04 — Who’s this guy

I’ve been building since 2009 — startups, companies, the ops software people actually touch. Agency years in the middle, then on my own in Austin. The hat is not a bit. Neither is the work. I embed, I ship, and I build AI that stays out of the way — the thesis, not a sticker on a marketing site.

Track record

  • 2009 — now

    Building

    Startups, companies, and the software that runs them.

  • 2015 — 2019

    At an agency

    Same job, more meetings.

  • 2019 — now

    On my own

    One person, no account manager. DoorCheck along the way.

House rules

  • I start with the work that’s eating your week — not a deck about “AI strategy.”
  • I won’t make your team prompt a robot for numbers they already needed on screen.
  • I won’t hand you a plan I can’t build myself.
  • I won’t sell you SEO and call it AI. That’s snake oil.

Proof

Real companies, real AI transformation — not a slide deck.

Case study · HelloFrom

A postcard company. A guest scans a code at the counter, writes a note on their phone, and HelloFrom prints it and mails it. Nothing for the place to stock.

Ocean shoreline at sunset
The front. Their own place, their own photo.

Hi — we made you a postcard. That’s your place on the front. Scan the code and it’s yours to keep sending.

— HelloFrom

claim your page

The back. Written, laid out, and addressed by an agent.

How I helped

The postcards were already working. They needed the next account without another hire. I wired agents into prospecting — find the place, write the note, style the card — invisible work, not another tool to log into.

  1. 01

    Finds the place

    Research and photos. No spreadsheet, no list to work through.

  2. 02

    Writes the note

    First person, in their voice, about that one specific place.

  3. 03

    Styles the card

    Photo, layout, the look of the thing. No designer in the queue.

  4. Then

    A person hits send.

    That part stays human. On purpose.

hellofrom.to

Their team still approves. The machine does the first pass.

Case study · FirstToSite

FirstToSite Austin

New permits, alerted to you first.

Every Austin municipal permit, scored by trade, scope, and scale. A contractor sets an alert. The match arrives. They reach the job before the crowd.

Crew on a jobsite in hard hats

Austin · last 14 days

The job. Two thousand permits a fortnight. The work is picking the right one.

Sep 18MatchResidential

8204 Wexford Dr

General / CarpentryComponent replacement

Job scale · 25 / 100

Alert sent · 2 min after filing

The alert. Jev classified it the moment it filed. Nobody asked a chatbot.

How I helped

They already had the permit feed. They needed the right job in a contractor’s pocket, not a spreadsheet hunt. I helped them plug in Jev — trade, scope, scale the moment each filing lands. No prompt. No chat widget.

  1. 01

    Reads the permit

    Every Austin filing, the morning it drops. No one is hunting the city’s website.

  2. 02

    Jev classifies it

    Instant. Trade, scope, scale. A roof is not a remodel. No one typed a prompt.

  3. 03

    Sends the alert

    Email or text. Address, what it is, how big. Already in their pocket.

  4. Then

    They call first.

    That part stays human. On purpose.

Next with them: automated prospecting — same invisible stack, pointed at who’s worth calling before the crowd shows up.

austin.firsttosite.com

Jev scores the permit. The contractor gets the match.

Case study · SchoolCheck

SchoolCheck

Checks-first for K-12. Know the round happened.

Doors, AEDs, extinguishers, emergency phones — checks for the morning walk, inspections when leaders need audit-ready depth. When something fails, tags and routing should already be right from what staff wrote.

School building exterior

Campus · checks today

The round. Main entrance passed. Gym AED inspected. Then the gate didn’t.

Playground gate latchFinding opened

Gate latch won’t hold shut

LatchGateLikely · hardware

Jev classified tags · 8:17 AM

The tags. From the issue summary — not a chatbot interview.

How I helped

They run safety operations across every campus. I helped them put Jev on finding classification — issue summary in, catalog tags out — plus the hooks into notifications and records so nobody re-labels the same latch in a spreadsheet.

  1. 01

    Run the check

    One question at the door, AED, or gate. Seconds — finish before the bell.

  2. 02

    Jev tags it

    Issue summary → the district’s tag catalog. Phrase match if Jev is offline.

  3. 03

    Record sticks

    Open finding, notifications, audit trail — every asset type, same platform.

  4. Then

    Staff fix it.

    That part stays human. On purpose.

schoolcheck.app

One tap on the check. Tags on the finding. Proof for the district.