AI Is Replacing the Interface, Not the Infrastructure

What the .com era taught me about AI and supply chain software

 

I love AI. It’s exhausting and exhilarating, and I haven’t felt this way about technology since 2000.

I joined my first startup, NetClerk, in 1997 as employee #3. We raised money from Granite Capital and Adobe Ventures and grew from three people to 300 in a matter of months. NetClerk helped specialty trade contractors – the electricians, roofers, and plumbers of the world – file for permits and business licenses online. It was a fragmented ecosystem, and we were trying to connect it through a medium that was still figuring itself out.

We were an ASP back then, what we’d call “early SaaS” today, and everything was 100x the work. We fought bandwidth limits. We built redundancy by hand. Browsers changed under our feet, coding languages came and went, and we worked through trial and error until the internet became reliable, browsers became stable and predictable, and backups became automated.

The last year of AI has felt exactly like that. New models, new capabilities, and new delivery mechanisms arrive almost weekly. The difference is speed. Distribution and operationalization have already stabilized. AI models keep getting better the way bandwidth did in 2000, but we already know how to distribute them, implement them, and harness them. What took the internet years took AI mere months. Since NetClerk, I’ve spent more than 20 years building Agistix through every technology shift that followed. None has moved this fast.

The Moments It Clicked

I had two “aha” moments that changed how I think about our business.

The first was watching ChatGPT extract shipment data from a photo of a shipping label with 99% accuracy. A picture taken on a phone became structured shipment data – mapped to our APIs and created a shipment in our platform.

The second was watching Claude pick up a code defect flagged in SonarQube, write a ticket, fix the defect, and check in the change for human review. It handled the whole loop, then handed it back to a person to approve.

That last step matters, and I’ll come back to it.

AI Is Headless Software

From a user experience perspective, AI is making software headless. Headless software separates the interface from the system that stores data, runs logic, and manages integrations, and AI is becoming that interface. In less than a year, everyone has grown accustomed to working this way. You attach a picture. You speak a request. You type a rough, qualitative description of what you need, and AI does the work.

In supply chain, this has translated to shippers, carriers, forwarders, brokers, suppliers, and customers all wanting the same thing: to type a request and instantly get an answer, a booked shipment, a tracking status, or an invoice paid without opening another application.

That’s a huge time saver. But the interface disappearing doesn’t mean the system disappeared. Behind every simple request, you still need the infrastructure to run it, the engineers who understand what AI is actually doing, the guardrails to secure it, and the resources to scale it. The simpler the front end gets, the more the back end has to carry.

We’ve Seen This Story Before

Financial markets have punished SaaS companies this year on the theory that AI will replace them. I’ve seen this movie, and the best preview is media.

The internet didn’t make content worthless. It made distribution cheap. Newspapers, radio, and TV stations used to own the expensive part: printing presses, broadcast licenses, delivery trucks. The internet cut out the middleman, and audiences suddenly had more choices about where to get content and who to pay for it.

Time Warner was the blue chip of that era, and it got the moment badly wrong. In 2000 it merged with AOL in a deal valued at roughly $165 billion. Two years later, AOL Time Warner reported a net loss of nearly $99 billion, at the time the largest annual loss ever reported by a public company, driven largely by write-downs on the value of the deal. Time Warner didn’t ignore the internet. It tried to buy its way in, too late, instead of rebuilding how it created and delivered content.

Other media companies moved early. Media General merged its Tampa newspaper, TV station, and website into a single newsroom, so one team could gather the news once and publish it everywhere. The Tribune Company merged with Times Mirror on the same idea: centralize content and distribute it efficiently across newspapers, TV, and radio. Not every early mover came through unscathed, but the companies that reorganized early learned to operate efficiently. That made them agile enough to pivot as bandwidth improved, cloud computing matured, and distribution kept changing, and to keep making money from their core asset: content.

And content never became free. People still pay for the sources they trust.

AI Is Doing to Expertise What the Internet Did to Distribution

The same shift is happening now across design, creative, legal, analytics, and software development. Subject matter expertise is no longer a rare commodity that commands a premium just for being scarce. AI has given everyone access to knowledge, the tools to create with it, and the means to deploy and manage what they create.

That doesn’t mean we need fewer experts. It means we need less of their time, and we need them overseeing the AI. It doesn’t mean we need fewer developers, either. It means all of us need new skills to work alongside AI to cut costs, raise quality, and deliver faster.

I see it every day at Agistix. Everyone here uses AI. We use it across the entire development lifecycle: coding, fixing, reviewing, standardizing, designing, and securing our infrastructure. We run it on our data warehouse for analytics and run agents that monitor transactional data around the clock.

The results are real. Our developers spend less time writing code and more time reviewing it. We’re delivering five times the story points per sprint and building microsites ten times faster. And every new hire at Agistix will earn a Claude certification, from Associate Foundations to Architect Professional depending on the role, and will be expected to build and manage agents. That’s the new baseline.

There Will Be Casualties

Software is heading for the same split media went through, and supply chain has a lot of legacy technology. Some software companies have spent years managing margins instead of paying down technical debt. Platforms stitched together through acquisitions, with fragmented codebases and disconnected ecosystems, will struggle to deploy AI across their portfolios. They’ll have to fix their foundations before they can accelerate, and some won’t make it in time.

Others have their heads down. The markets have hammered Salesforce and ServiceNow this year, but both have been busy building AI into their platforms rather than bolting it on. That’s what we’ve been doing at Agistix, too.

For supply chain technology, the core asset isn’t content. It’s our customers’ shipment data. The companies that win will keep evolving their development lifecycle and their infrastructure so they can deploy AI inside their platforms and on top of them. That’s how shipment data turns into streamlined operations, real savings, better collaboration, and stronger compliance.

Build vs. Buy in the Age of AI

AI is changing the buyer’s side, too. Customers can now run deep competitive analysis in an afternoon. Bad software used to survive on uninformed buyers, and that’s getting much harder. It also keeps pricing competitive, and that’s healthy for everyone. I welcome it. When buyers can check every claim, the platforms that do what they say they do win.

Of course, the same AI tools let companies attempt more on their own, and many are weighing whether to invest in technology or build it themselves. I understand the temptation. When a developer can generate working integration code in an afternoon, building in-house looks fast and inexpensive.

My caution is specific: don’t try to build transaction-based supply chain infrastructure because the first version looks easy. Think about what sits underneath a single shipment: orders, rating, bidding, pickups, booking, document automation, tracking, and invoicing. Every one of those constructs varies by mode, service, industry, and region of the world, and all of it runs across a fragmented ecosystem of data sources, each with its own schemas and protocols, and each changing on its own schedule.

AI speeds up how fast a team writes code, but it doesn’t reduce the number of data sources you depend on, slow down how often their schemas change, or lower the audit standard your data has to meet. For our customers in biotech, pharma, aerospace and defense, high-tech manufacturing, and utilities, that standard is high, because every shipment record is also a compliance record. 

At the end of the day, it comes down to this: launching a system takes weeks, but operating it efficiently takes years.

Simple on the Surface, Always Moving Underneath

Here’s what operating it looks like. Agistix connects to more than 4,000 data sources: carriers, forwarders, brokers, ERPs, compliance providers, and devices. Most release major updates every six to twelve months, and the list keeps growing. We’ve recently added localized fuel calculations, consolidation APIs, new denied-party screening providers, and IoT devices like temperature sensors and GPS trackers. Carriers are rolling out new pickup and proof-of-delivery APIs across every mode. On any given day, something underneath a shipment record is changing.

A shipment record isn’t a single feed. It pulls together the order, carrier events, sensor readings, compliance checks, documents, and invoice from many different sources. Our platform adopts each partner’s latest capabilities as they release them, without disrupting the simple experience users expect. When a carrier releases a new API version, the person asking their AI assistant for a tracking update gets the same answer, the same way, and never knows anything changed.

That flexibility comes from years of iteration. Every carrier change, new mode, and customer requirement has shaped how the platform handles the next one. And it keeps paying off: between our architecture and AI-assisted development, we’ve cut integration time by 50%, and we expect to keep driving integration time and cost down for our clients.

Our Vision: Any AI, Any Employee, One Trusted Platform 

Is AI replacing platforms like ours? Not yet, and I’ll be honest about the “yet.” AI-first competitors are coming, and they’ll push all of us. What AI is changing right now is how people engage with our software, and what they expect it to do: operationalize the work and surface new, deeper insights.

We’re building for that. Agistix already creates shipments from PDFs, images, and voice-to-text, configures agents around each customer’s supply chain, and draws insight from tens of billions of transaction records. Next year, we plan to take the next step: Model Context Protocol (MCP) infrastructure that connects Agistix directly to the AI our clients already use.

Think of MCP as a secure, universal adapter between AI assistants and the systems a company runs on. Any authorized employee will be able to work with Agistix in plain language from Claude, ChatGPT, Gemini, or whatever AI comes next. They’ll be able to create shipments, orders, and invoices. They can check status, report an exception, get quotes, pull documents, or forecast spend. All of it happens in the assistant they already have open.

Picture a salesperson about to call a customer. Today, she emails the logistics team and waits. With MCP, she asks her AI assistant for the status and proof of delivery on that customer’s shipment and has both in seconds, without ever logging in to Agistix.

Opening Agistix to any AI assistant only works if the controls stay just as strict. Every request will run through the same governed platform our clients rely on today, and access will be tied to each person’s company login, so employees see and act on only the data they’re authorized to see. The interface gets simpler. The controls underneath don’t get any looser.

Humans Must Stay in the Loop

AI gets things wrong. Anyone who uses it every day knows that. That’s why we believe every AI task needs human oversight, from a code fix to a shipment created from a PDF. Remember that Claude handed its fix back for human review. That isn’t a limitation of the workflow. It is essential to the workflow.

Security is what I think about most. Fortunately, we haven’t experienced an AI security incident in our platform, but the headlines are full of rogue AI, compromised models, and malicious code slipped into public AI repositories. Speed without guardrails is just risk moving faster.

What I’d Tell Every Supply Chain Leader

Use AI everywhere. Put it in front of your team, your partners, and your customers. Let it read the labels, draft the emails, fix the code, and surface the insights.

Just don’t expect AI to build your supply chain platform. Partner with people who have spent decades making the transactional layer correlated, secure, and reliable, so your AI has something trustworthy to work with.

I’ve been building technology to solve real-world problems my whole career, and I’ve never been more excited about what’s possible. AI is replacing the interface, not the infrastructure. The less you notice it, the better we’re doing our job.

If you’re weighing build versus partner, or want to see what your shipment data could do with the AI your team already uses, I’d love to compare notes. Grab time on my calendar.