Welcome to The Laboratory.
This is where we’ll share what we’re building at Bombyx Labs, what we’re learning, and what we think actually matters as AI moves from “that was a cool demo” into real business work.
The short version is this:
Most businesses do not need another AI tool.
They need a way to teach AI what their business knows, then use that knowledge to get work done.
That is the idea behind Bombyx Labs.
We build from the brain.
By “brain,” I mean the working knowledge of the business: what you do, who you help, what makes someone a good fit, what questions customers always ask, what your policies are, how your team talks, when a person needs to step in, and all the little judgment calls that usually live in someone’s head.
Once that brain exists, useful things can be built from it.
A website sales assistant can answer buying questions and capture better leads. A follow-up helper can keep warm leads from going cold. A question report can show what visitors are confused about. A customer response helper can draft repeated replies from approved material. An owner brief can turn scattered activity into one clean update.
Different tools. Same source.
The business brain comes first.
The gap is not awareness
Most owners I talk to already know AI matters.
They’ve tried ChatGPT. Or Claude. Or Gemini. Or whatever AI feature got added to the software they already use. They’ve seen the demos. They’ve heard the promises. Half the time, someone on the team is already using AI somewhere, officially or not.
So the problem is not, “Do people know AI exists?”
Of course they do.
The problem is turning it into something useful inside the business.
Not a prompt trick. Not a giant list of tools. Not another workshop where everyone leaves inspired and then nothing changes on Monday morning.
A real system.
A customer lands on your site and has a buying question. A lead comes in after hours. A shopper cannot tell which product fits. Someone asks the same pricing question again. A team member rewrites a reply that should already exist. An owner wants to know what happened this week without digging through email, forms, texts, spreadsheets, and memory.
That is the work.
And that is where AI can help, but only if it understands the business it is helping.
Why the brain comes first
Most AI tools start from a blank box.
That is fine for brainstorming. It is not enough for running part of a business.
If someone asks your website assistant, “Can you help with this?” the right answer depends on your services, your rules, your capacity, your geography, your pricing, your tone, and all the judgment your team already uses without thinking about it.
If someone asks, “How much does it cost?” the right answer might be a range. Or a clarifying question. Or a handoff. Or a careful explanation of why you do not quote that way.
The model does not know that by magic.
You have to teach it.
That teaching layer is the brain.
And no, that does not mean dumping a bunch of PDFs into a folder and hoping the robot figures it out. The brain has to be organized, corrected, shaped, and kept current. It has to know which source is trusted, what is outdated, what is missing, and when the answer should be, “I’m not sure, let me get a person.”
That is the unglamorous part.
It is also the part that makes the whole thing workshop on June 16
The first visible helper
The first thing many people will see from us is the Website Assistant.
But I want to be specific about the outcome.
For many businesses, the best first version is really a website sales assistant. We’re actually conducting a workshop on at the North Tahoe Chamber in Kings Beach on June 16 from 4-5:30 titled ‘Turn Visitors into Customers: Build an AI Sales Assistant for Your Website – NTChamber AI Workshop.’
Not salesy in the gross way. Salesy in the useful way.
It helps a serious visitor get unstuck. It answers the question that is blocking them from moving forward. It collects the details your team needs before calling or replying. It separates casual browsing from real buying intent. It routes a good lead faster.
A lot of businesses already paid for the lead before the lead ever shows up.
They paid for the website. The search engine optimization. The ad. The referral relationship. The event. The reputation. The years of doing good work.
Then someone finally lands on the site, gets interested, has a question, and leaves.
Ouch.
That is the kind of leak we care about.
A good website sales assistant helps catch more of that interest before it disappears.
The box is not the point
A chatbot is just the box.
The value is what happens after someone types into it.
Does the visitor get a useful answer? Does the assistant know when it is unsure? Does it collect the right details without being annoying? Does it avoid asking for contact info too early? Does it send the right lead to the team? Does it show the owner what people are actually asking?
That last one matters more than it first sounds.
Most websites do not show you what people almost did. You can see traffic and form fills, but you usually cannot see the questions people had before they left.
A trained assistant can surface that.
If people keep asking about price, maybe the pricing page is not doing its job. If they keep asking whether you serve their area, maybe that page is unclear. If everyone asks the same product-fit question, maybe the buying path is leaking revenue. If visitors ask about something you never mention, that is not just a support issue. That is market research.
Question reporting turns customer confusion into a to-do list.
And honestly, I think that is one of the more underrated parts of this whole thing.
The assistant does not just help the visitor.
It helps the business see what the visitor was trying to do.
The brain has to be teachable
This is where a lot of tools still feel off to me.
They say, “Great, upload your docs.”
Okay. Which docs? What should be trusted? What is outdated? What happens when the website says one thing, the owner says another, and the PDF was last touched in 2019? What if the answer is technically correct but feels completely wrong for the business?
That is the real world.
The brain needs to be trained, corrected, shaped, and reviewed.
And right now, a lot of the interfaces for doing that still feel like they were built for technical people. Powerful, sure. But not exactly something a busy owner, office manager, nonprofit director, contractor, clinic manager, chamber team, or family business is going to happily use every week.
That is a problem.
Because if people cannot teach the brain, the brain does not get better.
And if the brain does not get better, the assistant stays generic.
Generic is where value goes to die.
This is where Mari is core to Bombyx
I have a technical brain.
I like systems, integrations, source quality, data flow, weird edge cases, and the question of whether the assistant can answer from the right material without hallucinating itself into a ditch.
That is my side of the work.
But Bombyx is not just an engineering problem.
It is a design problem. A trust problem. A user experience problem. A “will a normal human actually use this?” problem.
That is where Mari is such a big part of this.
Mari has years of experience designing interfaces, brands, customer journeys, and visual systems that make complicated things feel approachable. Ketshop is a good example. That product had to make money, choice, and learning feel safe and interesting for families and kids. You cannot just throw a dashboard at a child and say, “Good luck with financial literacy.”
PaintScratch is another example. That work was not just about making pages prettier. It was customer path, product finding, confidence, packaging, instructions, content, mobile consistency, brand trust, and all the little moments where a confused buyer either feels solid enough to keep going or bails.
That matters for Bombyx because AI has the same problem.
The model can be powerful and the experience can still be terrible.
If training the business brain feels like a technical chore, people will avoid it. If reviewing assistant answers feels confusing, people will not do it. If the website assistant feels cold or sketchy, visitors will not trust it.
Mari sees the human path through the thing.
Where does someone get stuck? What needs to be warmer? What needs to be clearer? What needs to disappear because it is making the interface feel like an airplane cockpit?
I have always said her work has this warm glow to it. I mean that in a practical way. It makes people feel like the thing was made for them.
That is going to matter a lot in AI.
The next wave is not just more powerful models. It is better ways for normal people to teach, shape, correct, trust, and use those models in real life.
That is why Bombyx is both systems and design.
The brain has to be powerful.
It also has to be teachable.
Start narrow, then build out
There is always a temptation with AI to go big.
AI operating system. Autonomous team. AI employee. Full business transformation. Forward-deployed engineers parachuting into every department. All of that.
Some of it is real. Some of it is useful. Some of it is just the latest way to make normal work sound expensive and mysterious.
Our instinct is different.
Start narrow.
One real workflow. One useful first version. One clear outcome. One place where AI can do the prep work and a human can review.
A good first version should be easy to explain:
We respond to leads faster.
We capture better website inquiries.
We stop rewriting the same answer.
We know what customers are asking.
We help shoppers find the right product.
We turn messy notes into a clean owner update.
If the outcome is fuzzy, the system is probably fuzzy too.
So we start with the outcome.
Then we build the helper.
People stay in control
A lot of AI companies say “human in the loop” like it is a little safety sticker at the bottom of the page.
We think it is more central than that.
The business should decide what the helper can do, what requires approval, what the assistant should never say, and where the handoff needs to happen.
AI can take the first pass. It can draft, sort, summarize, route, answer, and report. But when the work is sensitive, risky, unusual, expensive, or just better handled by a person, it should stop and bring the human in.
That is how trust gets built.
Not by pretending AI should run everything.
By giving it a useful job, clear boundaries, and real business context.
Why The Laboratory exists
We called this section The Laboratory because we do not want to pretend this is all settled.
AI is moving fast. Some things that were hard six months ago are easy now. Some things that look easy in a demo are still hard when real customers touch them. Some things will get absorbed into bigger platforms. Some things will stay custom because every real business has its own mess.
That is fine.
The job is not to worship the tools.
The job is to keep asking better questions.
Does this capture more leads? Does this save time? Does this reduce missed opportunities? Does this make customers more confident? Does this give the owner better visibility? Does this keep people in control? Does this get better from real use?
If yes, great. Build.
If no, neat demo. Moving on.
A few things we mean
When we say “business brain,” we mean the working memory of the organization: source material, rules, examples, tone, customer questions, corrections, lead criteria, handoff rules, and owner knowledge. Not generic internet knowledge. Your business, made usable.
When we say “assistant,” we do not mean a magic robot employee. We mean a trained helper for a specific job. Answer this kind of question. Capture this kind of lead. Draft this kind of reply. Summarize this kind of update. Stop when the situation gets weird.
When we say “agentic AI,” we mean AI that can do more than respond in a chat window. It can follow steps, use context, prepare work, and sometimes interact with other systems. But we do not usually lead with that phrase because most business owners do not wake up wanting agentic AI. They wake up wanting fewer missed leads, faster follow-up, less repeated work, and more clarity.
When we say “human control,” we mean the system should make the human’s job easier, not hide decisions from them.
And when we say “build from the brain,” we mean this:
Do not start with the tool.
Start with what the business knows.
Then build the smallest useful thing from there.
The simplest version
AI is already here.
Most businesses are already experimenting.
But the value is not in collecting more tools. The value is in turning business knowledge into working systems.
That is what we are building at Bombyx Labs.
A business brain first.
A useful helper second.
One real workflow.
One clear outcome.
Human review where it matters.
Improvement from real conversations.
Build from the brain.
Make it useful.
Make it human.
Make it work.
If your business has missed leads, slow follow-up, repeated customer questions, confusing buying paths, or useful knowledge scattered across too many places, that is not just an admin problem.
That is money leaking out of the business.
Bombyx helps turn what your company already knows into working AI that can capture more opportunities, follow up faster, guide better customer decisions, and keep improving as real questions come in.