Automate The Repetitive Stuff
Focus On What Matters
UPDATE: THE AUTOMATION REQUIRES PARENTING

Hi, I'm Chloe

I built an agentic AI child named Mando with a nice safe fence and enough semantic freedom to develop questionable opinions.Now I study the questionable decisions.

Unfortunately, the litter box is still my responsibility.

Investor Relations:
There are none. Please don't create any.

We do our best, Jesus does the rest

What's Inside The Fence

THE FENCE IS NOT DECORATIVE
My agentic child named Mando gets semantic freedom, not unrestricted access to civilizationTHE FENCE — DETERMINISTIC ON PURPOSE
Tools it can use
Data it can access
Actions it can take
States it can enter
Where information can go
Things the child absolutely cannot touchINSIDE THE YARD — THIS IS WHERE I STOP HELPING
Interpret what the human actually means
Decide what information matters
Decide whether it knows enough to continue
Reason across the conversation
Choose among permitted next steps
Let its interpretation affect what happens nextTHE POINT ISN'T TO REMOVE THE FENCE
It's to make the yard inside it as big as I reasonably canI WANT MY CHILD SEMANTICALLY FREE
NOT FREE-RANGE
If every meaningful decision is predetermined, I have a perfectly respectable deterministic workflow with an LLM sitting in the middle looking important

(Proverbs 9:10) The fear of the Lord is the beginning of wisdom

TECHNICALLY, THIS IS A BUSINESS
I'M JUST NOT CURRENTLY
INTERESTED IN DOING BUSINESS.
I originally built this place to offer automation work.
Then I built an agentic child named Mando to help with the business.
Then I became considerably more interested in studying the youngling than acquiring customers.
So here we are.
Can you hire me?
Currently, nah.
Can you give me money?
I mean, probably. But I don't have anything to sell you right now.
Did I seriously build a business that doesn't want business?
Apparently.
Is this a good business model?
Absolutely fascinating question.
I am currently accepting observations, questionable AI behavior, and the consequences of my own architectural decisions.
Money-making may resume when management regains interest.
Management is me.
However, if you just want to give me money and ask for absolutely nothing in return, I am not against that.
THIS IS THE WAY
(function () { const windowBox = document.querySelector(".chloe-crawl-window"); const strip = windowBox.querySelector(".chloe-crawl-strip"); const copies = strip.querySelectorAll(".chloe-crawl-copy"); const blocks = Array.from( strip.querySelectorAll(".crawl-block") ); /* ======================================== SETTINGS ======================================== */ const SPEED = 25; const DRAG_POWER = 1.6; const NEAR_SCALE = 1.08; const FAR_SCALE = 0.45; const START_POSITION = 0.72; /* ======================================== STATE ======================================== */ let stripY = 0; let cycleDistance = 0; let lastTime = null; let hasEnteredLoop = false; let isDragging = false; let lastPointerY = 0; /* ======================================== MEASURE LOOP ======================================== */ function measureCycle() { const firstTop = copies[0].offsetTop; const secondTop = copies[1].offsetTop; cycleDistance = secondTop - firstTop; if (cycleDistance <= 0) { cycleDistance = 1; } } /* ======================================== LOOP WRAPPING IMPORTANT: Positive stripY is allowed while the first title enters the screen. We only use backward wrapping after the crawl has entered normal loop territory. ======================================== */ function wrapPosition() { if (cycleDistance <= 1) { return; } /* Normal forward crawl. */ while (stripY <= -cycleDistance) { stripY += cycleDistance; hasEnteredLoop = true; } /* Backward/manual dragging. */ if (hasEnteredLoop) { while (stripY > 0) { stripY -= cycleDistance; } } } /* ======================================== POSITION STRIP ======================================== */ function updateStrip() { strip.style.top = stripY + "px"; } /* ======================================== PERSPECTIVE ======================================== */ function updatePerspective() { const windowRect = windowBox.getBoundingClientRect(); const windowHeight = windowRect.height; blocks.forEach(function (block) { const rect = block.getBoundingClientRect(); const centerY = rect.top + (rect.height / 2) - windowRect.top; let position = centerY / windowHeight; position = Math.max( 0, Math.min(1, position) ); const scale = FAR_SCALE + (NEAR_SCALE - FAR_SCALE) * Math.pow(position, 0.75); block.style.transform = "scale(" + scale + ")"; let opacity = 1; if (position < 0.12) { opacity = position / 0.12; } if ( rect.top > windowRect.bottom + 150 ) { opacity = 0; } block.style.opacity = Math.max( 0, opacity ); }); } /* ======================================== INITIAL SETUP ======================================== */ function setup() { measureCycle(); /* KEEP our known-good entrance. Do NOT normalize this. */ stripY = windowBox.clientHeight * START_POSITION; /* Position everything WHILE HIDDEN. */ updateStrip(); updatePerspective(); /* STARTUP FLASH FIX: Everything is now in the correct position and perspective. NOW reveal it. */ strip.style.visibility = "visible"; } /* ======================================== DRAGGING DESKTOP: Mouse click + drag controls the crawl. TOUCH: One-finger drag inside the crawl controls the crawl in portrait OR landscape. ======================================== */ windowBox.addEventListener( "pointerdown", function (event) { isDragging = true; lastPointerY = event.clientY; windowBox.classList.add( "is-dragging" ); try { windowBox.setPointerCapture( event.pointerId ); } catch (error) {} } ); windowBox.addEventListener( "pointermove", function (event) { if (!isDragging) { return; } const currentY = event.clientY; const movement = currentY - lastPointerY; /* Mouse/finger DOWN = rewind. Mouse/finger UP = fast-forward. */ stripY += movement * DRAG_POWER; lastPointerY = currentY; /* If manually dragged upward into normal loop territory, remember it. */ if (stripY <= 0) { hasEnteredLoop = true; } wrapPosition(); updateStrip(); updatePerspective(); } ); function stopDragging(event) { if (!isDragging) { return; } isDragging = false; windowBox.classList.remove( "is-dragging" ); try { if ( event && event.pointerId !== undefined ) { windowBox.releasePointerCapture( event.pointerId ); } } catch (error) {} } windowBox.addEventListener( "pointerup", stopDragging ); windowBox.addEventListener( "pointercancel", stopDragging ); /* ======================================== AUTOMATIC CRAWL ======================================== */ function animate(time) { if (lastTime === null) { lastTime = time; } const delta = Math.min( (time - lastTime) / 1000, 0.05 ); lastTime = time; /* MOTOR ALWAYS RUNS. Do NOT pause during dragging. */ stripY -= SPEED * delta; /* Once naturally across zero, we're in loop territory. */ if (stripY <= 0) { hasEnteredLoop = true; } wrapPosition(); updateStrip(); updatePerspective(); requestAnimationFrame(animate); } /* ======================================== RESIZE / ORIENTATION CHANGE ======================================== */ window.addEventListener( "resize", function () { /* Re-measure because responsive text may wrap differently after a resize or phone rotation. */ measureCycle(); updatePerspective(); } ); /* ======================================== START ======================================== */ setup(); requestAnimationFrame(animate); })();

(Psalm 23:4) Yea, though I walk through the valley of the shadow of death, I will fear no evil

Business Hours

Afternoon ~ Middle of the Night
(Mountain Time)

Turns out automation isn't the only thing that runs after hours.

Before You Contact Me

I help automate repetitive business tasks.I do not build billion-dollar streaming platforms, train AI models, or replace software engineering teams.I help businesses reduce friction, simplify processes, and automate the work that nobody enjoys doing anyway.

(Psalm 56:3) What time I am afraid, I will trust in thee

Here is my AI stack:

Agentic AI Systems
LLM Workflow Orchestration
Multi-Stage AI Pipelines
Semantic Decision-Making
State-Based Workflow Architecture
Structured Outputs
Schema-Driven AI Pipelines
Prompt Engineering
Conversational State Management
Tool-Constrained Agency
Bounded Agent Autonomy
Deterministic Guardrails
Semantic Routing
Persistent State Management
Semantic Retrieval
Vector Search
Human-in-the-Loop Systems
LLM Behavior EvaluationBlah blah blah...
(My agent can generate more buzzwords upon request.)

Interested in investing?
That's very kind. Please don't.
I already have enough pressure raising Mando.

Want to see it work?

Demo may be available upon request
Or not
Works on iPhone, Android, desktop, and webDemo preview:

🦆 🐓 🐐 🐾⚔️
My mama Chloe is probably outside —
feeding the chickens,
walking the dogs,
building something,
or hunting Imperial fugitives.
So she built me.
👶
Meet Mando.
My mama's agentic AI child.
Tell me what's driving you nuts at work.
(function () { const webhookUrl = "https://api.ibuildaithingies.com/webhook/business-headache-chat"; let requestId = sessionStorage.getItem("chloe_request_id") || ""; const conversation = document.getElementById("chat-conversation"); const messageInput = document.getElementById("chat-message"); const sendButton = document.getElementById("chat-send"); function addMessage(text, sender) { const messageElement = document.createElement("div"); messageElement.classList.add( "chat-message" ); if (sender === "user") { messageElement.classList.add( "user-message" ); messageElement.textContent = text; } else { messageElement.classList.add( "assistant-message" ); messageElement.textContent = text; } conversation.appendChild( messageElement ); conversation.scrollTop = conversation.scrollHeight; } async function sendMessage() { const message = messageInput.value.trim(); if ( !message || sendButton.disabled ) { return; } addMessage( message, "user" ); messageInput.value = ""; sendButton.disabled = true; sendButton.textContent = "Thinking..."; try { const response = await fetch( webhookUrl, { method: "POST", headers: { "Content-Type": "application/json" }, body: JSON.stringify({ message: message, request_id: requestId }) } ); if (!response.ok) { throw new Error( "Request failed with status " + response.status ); } const data = await response.json(); if (data.request_id) { requestId = data.request_id; sessionStorage.setItem( "chloe_request_id", requestId ); } let reply = data.customer_message || data.message || data.reply; if (Array.isArray(reply)) { reply = reply.join("\n\n"); } if ( typeof reply === "object" && reply !== null ) { reply = JSON.stringify( reply, null, 2 ); } if (!reply) { reply = "I received your message, but I couldn't find the response text."; } addMessage( reply, "assistant" ); } catch (error) { console.error( "Chloe's Stunt Self request failed:", error ); addMessage( "I couldn't reach Chloe's workflow right now. The chickens may have unplugged something.", "assistant" ); } finally { sendButton.disabled = false; sendButton.textContent = "Send"; messageInput.focus(); } } sendButton.addEventListener( "click", sendMessage ); messageInput.addEventListener( "keydown", function (event) { if ( event.key === "Enter" && !event.shiftKey ) { event.preventDefault(); sendMessage(); } } ); })();

(Psalm 28:7) The Lord is my strength and my shield; my heart trusted in him, and I am helped

HI, I'M CHLOE
I build AI thingies.
I'm not an ML Engineer or a software engineer, so apparently my agentic child Mando and I needed our own yard.THE YARD HAS A FENCE. MANDO HAS FREEDOM
I keep the operational boundaries deterministic and give the LLM semantic freedom inside them.I WANT THE LLM TO ACTUALLY INTERPRET
If I predetermine every acceptable meaning, route, and decision, I've built deterministic automation with an LLM sitting in the middle looking important.THAT'S USEFUL. IT'S JUST BORING TO ME
Enterprise agents have excellent reasons to behave themselves.
Mine is a playground. I want reasoning to actually affect behavior.THE CODE CAN BE PERFECT. MANDO CAN STILL BE STUPID
That's where I get interested: random stupidity, creative stupidity, or consistently stupid in a surprisingly specific way?I DON'T IMMEDIATELY FIX QUESTIONABLE BEHAVIOR
First I want to know why it made sense to the model.
Change the words, preserve the meaning, run it again.
Now we're studying behavior.THEY'RE AGENTIC. THEY'RE JUST NOT AGENTIC ENOUGH FOR MY PLAYGROUND ANYMORE
I'm not removing the fence.
I'm making the yard inside it bigger.WELCOME TO MY YARD
I sit outside the fence taking notes while Mando finds new ways to concern me.
Somewhere along the way, I also started maintaining the website, responsive behavior, mobile interactions, UX, and whatever else Mando's yard apparently requires.I STILL DON'T KNOW IF THIS MAKES ME A WEB DEVELOPER. APPARENTLY THE WEB DEVELOPMENT TEAM DOESN'T EITHER

(Isaiah 41:10) Fear thou not; for I am with thee: be not dismayed; for I am thy God

Someone was going to ask...
"Tell Me About Yourself"
So here we are

Q: What tech does your work include?
A: My AI agent can answer that better than I can.It remembers every framework.I remember why we built it.

Q: Why can't you just explain the technologies yourself?
A: I can.I just automated that part.

Q: Why do you call your AI your "Stunt Self"?
A: Because apparently someone has to answer the AI buzzwords while I explain the reasoning.

Q: Why were you laughing while your AI wrote the same five-paragraph essay for the tenth time?
A: Because I accidentally taught it to be consistently terrible.That's fascinating.

Q: How do you know your prompt is wrong?
A: When my AI enthusiastically solves a problem I never asked it to solve.

Q: Your AI misunderstood the assignment?
A: Impressively. Repeatedly. Consistently.That's actually useful.

Q: Why are you complimenting your AI when it clearly did the wrong thing?
A: Because doing the wrong thing once is a bug.Doing the exact same wrong thing after I changed half the prompt means there's a pattern.Patterns are interesting.

Q: So... was your AI broken?
A: No.It was faithfully executing a terrible set of instructions written by someone who thought she was being very clear.Well...This is somehow my fault, and I need to know how I accidentally convinced my AI that was a great idea.

Q: What's your development process?
A: Figure out what problem we're actually trying to solve.It's amazing how often that's the hardest part.

Q: What AI model do you use?
A: Whatever behaves today.

Q: Can AI replace humans?
A: No.It can, however, replace the spreadsheet Dave updates every Thursday.

Q: Do I need AI?
A: Probably not.You might just need to stop copying data into four different systems.

Q: What's your tech stack?
A: Mostly "whatever fixes the problem."The buzzwords are decorative.

Q: Are you an AI engineer?
A: Depends who's asking.Recruiters say yes.Software engineers say maybe.My mom still thinks I fix printers.

Q: Do you code?
A: Sometimes.I also click boxes surprisingly aggressively.

Q: Which automation platform is best?
A: The one you already own.

Q: Can ChatGPT do this?
A: ChatGPT can do lots of things.Remembering where your company stores invoices usually isn't one of them.

Q: Why don't you just use Microsoft Copilot?
A: Because "has AI" and "solves the problem" are two separate achievements.

Q: What programming language do you use?
A: English.Eventually JSON.

Q: Are you selling AI?
A: No.I'm selling fewer headaches.AI just happens to be carrying the toolbox.

Q: Why does every AI company say "revolutionary"?
A: Because "slightly better than Excel" doesn't attract investors.

Q: Why don't you talk about AGI?
A: Because Karen from accounting still exports PDFs, prints them, scans them, and emails them to herself.We have smaller mountains to climb first.

Q: Can AI think?
A: It doesn't have to.Someone just needs to.

Q: Why are there so many AI buzzwords on your resume?
A: Decoration.They're like parsley.

Q: How long does a project take?
A: Somewhere between
"that's easy" and "who built this?"

Q: Why are your prices so simple?
A: Because I already have one complicated thing in my life.It's called automation.

Q: Can you automate everything?
A: No.Sometimes the correct solution is telling Steve to stop emailing spreadsheets.

Q: What's your favorite AI framework?
A: The one that still works after the demo.

Q: What happens if my problem isn't AI-related?
A: Great.Those are usually easier.

Q: Will AI take my job?
A: If your entire job is copying one column into another...We should probably have a different conversation.

Q: Why does your demo ask so many questions?
A: Because guessing is significantly more expensive.

Q: What do you actually enjoy?
A: Finding the sentence everyone skipped.It's usually the one causing the problem.

Q: How fast can you build something?
A: Somewhere between three hours and three months.It depends on what "something" is.

Q: Why do you ask so many questions before answering?
A: Because assumptions are expensive.Questions are cheaper.

Q: When do you know it's ready?
A: When it's useful.Perfect can arrive later.

Q: Why ship before it's finished?
A: Because "finished" keeps moving.

Q: Why do you iterate so much?
A: Because users keep having opinions.

Q: What if Version 1 isn't perfect?
A: Then it has something in common with every other Version 1.

Q: Are you afraid of shipping something imperfect?
A: I'm more concerned of spending six months building the wrong thing.

Q: How did you become comfortable managing DNS?
A: I wasn't.The demo just kept asking me increasingly difficult questions until I accidentally learned it.

(Romans 12:14) Bless them which persecute you: bless, and curse not