Tell me what you're
working on.

Or what you're trying to solve. Or where you're stuck.

Thirty years across machine vision, motion control, IoT, cloud platforms, machine learning, and now AI. Cognex acquired my last company for $115M. These days I run an AI consulting practice for small and mid-sized businesses — and I build my own AI-run operation in the open, so you can check whether I practice what I sell. SMB founders bring me in when the problem isn't on a slide deck yet — when they just need to talk it out with someone who's seen this kind of thing before.

Start a conversation

What people bring me

"We're a small company and AI feels overwhelming."
Where to start, what to skip, what's actually going to ROI in six months.
"Our product needs to do something AI can probably do."
Architecture, vendor choice, build-vs-buy, what the right MVP actually looks like.
"We hired an AI consultant and they gave us a strategy deck. Now what?"
Translating decks into shipping work. The thing nobody on the team wants to own.
"We're a non-AI software company and our customers keep asking for AI features."
Roadmapping the real signals from the noise. What ships first.
"We don't have a tech problem. We have an operations problem."
Sometimes that's where AI helps; sometimes it isn't. I'll tell you straight.
"I'm not sure if this is even an AI problem."
Often it isn't. Twenty years of B2B sales operations and channel work taught me that most "tech" problems are people-and-incentives problems wearing a tech disguise.

I have done this across a lot of stacks.

1990s–2000s Machine vision sales + BD at DVT — incubator to the Cognex acquisition How to find the wedge customer before the market knows it wants you
2000s–2010s DataComplex (cloud event-processing). TrackPlus (mobile). DHC (ML/CV consulting) Most engineering problems are actually product problems
2010s–2020s LiteSheet, Mitsubishi, Candidus — industrial IoT, motion control, agricultural tech. Won 2021 Tech of the Year Hardware-software-cloud systems live or die by the boring integration work
2023–2026 Saivvi, CTO. Built AI tools for operators who use software but don't build it A good product doesn't survive rented distribution
Longer version

What I'm building right now

The practice

AI consulting for small and mid-sized businesses that win work through direct sales. Not strategy decks — running loops: I leave your team with a working process, wired into how you actually operate. /work

Emit & Compound — built in the open

The institutional-memory pattern I run my own content business on: every lesson an AI session learns becomes a rule every future session honors. I'm building and publishing the whole thing publicly — the pattern doc, the logs, the mistakes. /pattern

Writing

Essays and build logs from inside an AI-run business. What's working, what isn't, with the receipts. No tips, no listicles. /library

How this works

Email me. Tell me what you're working on. We get on a call.

If it's something I can help with — a single conversation, a short project, a monthly retainer, an introduction — we figure out the shape that fits. If it isn't, I'll point you to someone better suited.

I take a small number of engagements at a time so the work is real, not slideware.

david [at] saivvi.com


See the engagement shapes directly

Saivvi

What I built, what I learned.

From August 2023 to 2026 I was CTO of Saivvi, an AI product company for non-technical operators — the marketing manager, the ops lead, the SMB founder. The bet: people who run businesses should be able to own an AI workflow, not just consume one somebody else built.

We built an iPhone app backed by AI workflows — the hard part was never the model. It was the orchestration: turning a draft idea or a shared link into polished, on-brand output a non-technical person would actually ship. Content production pipelines, brand-voice layers, the unglamorous glue.

How it ended

The product was good. Users told us so — the reviews praised the layout, the organization, and the content ideas it put in their hands. We were on the yellow brick road. What ended it wasn't the product: it was the price of reaching the next customer. Social-media advertising inflation pushed acquisition costs past what the economics could carry, and a good product you can't affordably put in front of people is a good product winding down. The app is still available; the company is closing its chapter.

Three lessons I carry into every engagement now

  1. A good product doesn't survive rented distribution. Our reach lived on ad platforms, and when their prices moved, our fate moved with them. It's why everything I build now runs on owned distribution — an audience, a library, a pattern shared openly — that no auction can reprice.
  2. The demo runs on capability; the finish runs on scaffolding. The model was never our bottleneck. The orchestration, the brand-voice layer, the pipeline glue — the unglamorous parts were the product. Budget for them first.
  3. Most "tech" problems are people-and-incentives problems wearing a tech disguise. Building for non-technical operators taught me where AI actually sticks in a business: not where the technology is impressive, but where someone's real workflow gets shorter.

That chapter is why the practice looks the way it does: I've been on the product side of the AI promises SMBs get sold. When I tell you what's real and what's theater, it's because I've built both.

What I do now

Here's exactly how
to work with me.

Five shapes of engagement. Rough pricing. What you should bring. What I won't do. Pick the one that fits — or email me and we'll figure it out.

Five shapes

1
A single strategy call
Two hours. We work through your specific problem. You leave with a recommendation, three concrete next steps, and a straight answer on what's worth pursuing.
Best for: "We need a sanity check before we commit."
$1,500 — pay after the call if it was useful.
2
A two-week sprint
Scoped engagement to ship one specific thing: an MVP, a vendor evaluation, an AI feasibility study, a roadmap that actually maps. Daily check-ins. Written deliverable at the end.
Best for: "We need this figured out and shipped, not researched."
$15,000 – $30,000 depending on scope.
3
A monthly advisor retainer
On-call for your team. Calls when you need them, Slack between. 5–10 hours a month. I'm the second opinion on architecture, product, hiring, vendor decisions, the hard conversations.
Best for: founders and CTOs who don't need a full hire but want someone senior in the loop.
$5,000 / month — three-month minimum.
4
Fractional Head of AI
10–25 hours a week embedded with your team. Strategy + builds + the awkward conversations. I sit in your standups, write the AI roadmap, evaluate vendors, code where it matters, and ship.
Best for: SMBs that have an AI strategy and need someone to run it.
$10,000 – $15,000 / month — six-month minimum.
5
Team workshop or training
Half-day or full-day session for your leadership team. What AI can and can't do for your specific business, how to evaluate vendors, how to think about build-vs-buy, what to ship first.
Best for: leadership teams that need everyone on the same page before committing budget.
$7,500 – $15,000 depending on prep and format.

Or — something else entirely.
Plenty of useful work doesn't fit a template. Email me with what you've got.

How we start

  1. Email me at david [at] saivvi.com with what you're trying to solve. A paragraph is plenty — don't write me a brief.
  2. 30-minute call within a week. Free. I'll tell you honestly whether I'm the right fit. If I'm not, I'll point you to someone who is.
  3. Pick a shape and start. One-page agreement, not a twelve-page MSA. We start the following week.

Who this is for, and who it isn't.

Good fit

SMBs and mid-market companies, roughly 10 to 500 people. Founders, CTOs, COOs. Teams that ship. People with a real problem who want to talk to a real practitioner about it.

Bad fit

Enterprises that need a fifty-page strategy deck before anyone is allowed to make a decision. Companies looking for someone to validate a choice they've already made. Stealth founders who can't describe their product until I sign an NDA. Anyone shopping for the cheapest consultant — I'm not it.

What I've shipped

DVT Corporation, Director of Business Development, 1992–2005
$115M acquisition by Cognex. Built the channel that took DVT from incubator to acquisition over thirteen years.
DataComplex, founder, 2005–2009
Cloud event-processing platform before "cloud" was a buyer category.
Candidus, VP of Engineering, 2019–2023
Hardware + software + cloud control system for commercial greenhouses. Won the 2021 Medal of Excellence — Technology of the Year.
Saivvi, CTO, 2023–present
AI product company for non-technical operators.

Plus thirty years of channel work, B2B sales operations, and consulting across motion control, machine vision, industrial IoT, agricultural technology, and computer vision.

Start here

david [at] saivvi.com

A paragraph about what you're working on is plenty.

The Library

Patterns, build logs, and the systems behind them. Everything I learn building an AI-run business, written down and shared — the way Ed and Helen would have.

1 published · more landing this week

00 · THE PATTERN · ~15 MIN READ
Emit & Compound: institutional memory for businesses that run on AI
The complete pattern — one shared ledger, a ten-minute weekly ritual, and a rules page every AI tool reads. Any stack, no code, start this week. With the fully-automated ceiling shown in real logs.

Get the pattern or get it by email

01 · BUILD LOG · ~4 MIN READ
The rule that evaporated
My AI invented a statistic. I caught it, wrote a rule, and watched the rule die in a folder nothing re-reads. The failure that started this whole build — and the series.
— lands this week
02 · SYSTEM · ~5 MIN READ
I don't hunt for things to post
A machine already read the library: 32 sources fanned into 190 interlinked pages, so mining a post is reading a map, not searching a pile. The content engine, shown from the inside.
— lands this week
03 · BUILD LOG · ~5 MIN READ
The correction that survived
The payoff: three corrections from one working session became permanent rules — a gate note, a reviewer nitpick, a redirect — and the very next run loaded all six. Real log lines, zero human carry.
— lands this week
04 · STACK · ~10 MIN READ
The SMB AI stack, layer by layer
The seven-layer breakdown of what an AI-running small business actually needs — what to buy, what to skip, what order to build in. The series the stack posts come from.
— lands this week
05 · SYNTHESIS · IN PROGRESS
Why AI projects don't finish
Ten mechanisms that stall AI projects between the demo and the finish line — compiled from sources across the wiki. Publishing as a series; the mechanisms land here as they ship.

Follow the series

If you want shorter and more frequent: I'm on LinkedIn and X.

Emit & Compound

What exactly is "the leak"?

Every business running on AI is learning things constantly — and losing almost all of it.

Someone corrects a bad AI draft. Someone works out the phrasing that gets replies. Someone catches the claim that would have caused a compliance headache. Each lesson lives in one person's chat history, one closed conversation, one head — and the next person, the next tool, the next week starts from zero. You're paying for the same lesson over and over. That's the leak.

The fix isn't a smarter model. It's a small process: one shared ledger where lessons land, a ten-minute weekly review where the owner approves the ones that matter, and a rules page every AI tool in the business reads before it works. Set it up once — with a spreadsheet and a doc, no code — and a correction made on a Tuesday binds every seat, every tool, permanently.

I run the fully-automated version of this every day (my rules file grew from three rules to six in a single working session, with nothing carried by hand). The pattern that makes it work is free below. If you'd rather have it wired into your business by Friday, that's a call.

One ledger row, so you can picture it

Jul 9 · AI draft review · claim · "Never quote a market stat without a named source — AI invented one that nearly shipped" · marketing · approved

Pick your leak

Sales
Your best rep's objection-handling line — the one that actually lands — lives in their head, and it leaves when they do. With the loop running: the line gets proposed to the ledger the week it's discovered, approved once, and every AI-drafted follow-up from every seat uses it from then on.
Support
A rep figures out the workaround for the billing glitch. The next rep rediscovers it from scratch — on a customer's time. With the loop running: the workaround is a status row the day it's found, and the AI answering support drafts already knows it.
Marketing
Your AI invents a statistic that sounds right. Someone catches it — this time. The rule that would stop the next one dies in that chat. With the loop running: "cite it or cut it" becomes a house rule every drafting session loads automatically. The catch happens once; the protection is permanent.
Operations
First-touch emails keep promising delivery dates ops can't always honor — because the person drafting them changed and nobody told the new one. With the loop running: the rule is in the rules page, the rules page is in every tool, and the promise stops going out — no matter who's drafting.

If it's a lesson someone learned and someone else will need, it belongs in the ledger. The full pattern — ledger, ritual, rules page, and the setup for your exact tools — is free.


How I work with businesses: /work

The Library: /library

About

The short version

I'm David Hicks. Thirty years of building things — machine vision, motion control, IoT, cloud platforms, machine learning, and now AI. Sold one company to Cognex for $115M. Lost some. Built more. These days I run an AI consulting practice for small and mid-sized businesses, and build my own AI-run operation in the open.

Based in Atlanta.

The longer version

I started in 1992 at DVT Corporation, a machine vision startup nobody had heard of. I was the Director of Business Development. We spent thirteen years finding our wedge customer, building the distribution channel, and learning what most consultants get wrong about industrial buyers. In May 2005, Cognex acquired DVT for $115 million.

I learned three things in those thirteen years that have shaped everything since:

  1. Most "tech" problems are people-and-incentives problems wearing a tech disguise.
  2. You find the wedge customer before the market knows it wants you.
  3. The boring integration work is what makes systems live or die.

Since DVT, I've built or led engineering at six more companies. DataComplex — a cloud event-processing platform in 2005, before "cloud" was a buyer category. TrackPlus, mobile-web apps for conferences and non-profits. DHC, machine learning and computer vision consulting through the late 2010s. Candidus, where I was VP of Engineering and led the build of a hardware-software-cloud control system for commercial greenhouses; we won the 2021 Medal of Excellence — Technology of the Year.

From August 2023 to 2026, I was CTO of Saivvi. AI was the next wave. I'd been working in machine learning and computer vision for years, so the technology wasn't new to me. What was new was the wedge audience: not engineers, but the operators who use software and don't build it. The marketing manager. The ops lead. The SMB founder. The product was good; the economics of reaching its customers weren't — the Saivvi chapter tells that story straight, including the three lessons I charge for now.

What hasn't changed

The technology has changed every decade. The job hasn't.

Find the customer who needs this. Listen to them. Build the smallest thing that works. Ship it. Listen again. Repeat.

Everything else is decoration.

What I do when I'm not working

I write. I read. I spend time with family. I think about what the right next thing is more than is probably healthy.


If you want to work together: /work

If you want to follow the build: /now

If you want the long-form: /library

Now

Rolling log of what I'm working on, learning, and shipping. Updated when the work warrants it — not on a posting schedule. Newest first.

July 2026

The practice

Full pivot from product (Saivvi, now winding down) to an AI consulting practice for SMBs. The thesis: AI is only worth anything when it's mapped onto your actual business process — I sell what it does to how you run and grow the business, not the tool.

Emit & Compound

Wrote the pattern document — institutional memory for businesses that run on AI: one shared ledger, a ten-minute weekly ritual, a rules page every tool reads. I run the fully-automated version daily; this month it started compounding for real (my rules file grew from three rules to six in a single working session, nothing carried by hand). A build-in-public series about it is queued to start publishing this week.

The content engine

The wiki that mines my posts crossed 32 sources fanned into 190 interlinked pages. Every post I publish comes out of it — the machine reads the library, I read the map.

Open question I'm sitting with

My knowledge system's biggest flaw is that it never forgets — pure accretion, no decay policy. Rules retire by hand today. What does governed forgetting look like when the ledger is five hundred rows deep?

Not doing

Cohort-based courses. Posting on a schedule instead of when the work warrants it. Anything that requires saying "leverage" or "transform" with a straight face.

May 2026

Saivvi

Working on the iPhone app's content production pipeline. Specifically: turning a draft idea or shared URL into a polished short-form reel ready to post. The hard part isn't the AI; it's the orchestration of the AI plus the visual production plus the brand-voice layer. Lots of Remotion. Lots of MCP-based tooling.

Reading

Working through the early literature on the autoresearch pattern (Karpathy's recent posts, davebcn87/pi-autoresearch). Thinking about how it applies to brand-awareness loops at the SMB scale.

Writing

Drafting an essay on the gap between "AI advice for developers" and "AI advice for everyone else." Why that gap keeps widening instead of closing, and what I think that means for the next 24 months of product opportunity.

Open question I'm sitting with

What does it look like for a non-technical operator to own their AI workflow — not just consume one someone else built — without having to learn to code? Saivvi has one answer. There are probably others.

Not doing

Cohort-based courses. Newsletters with more than one post a month. Anything that requires saying "leverage" or "transform" with a straight face.

Inspired by Derek Sivers' /now page movement.