AI Is Becoming a Commodity. Context Is Becoming the Moat.

In this session’s Office Hours, three founders showed up with variations of the same product: voice AI, automated chat widgets, AI CRM assistants. All three were technically competent. All three were hitting the same wall.

Ian Arden’s answer to each of them was the same, and it’s worth unpacking.

“AI itself is becoming the commodity. Text-to-speech, speech-to-text — those are problems that larger companies have already solved for us. Where the know-how and the IP lies is in this mid-layer of accruing all the knowledge base for your customers and empowering them to leverage AI as knowledgeably as possible.”

That mid-layer is context. And right now, most founders are building below it.

Why Every AI Founder Is Building the Same Thing

Open any AI startup directory. You’ll find hundreds of companies building voice agents for appointment booking, chatbots for website lead capture, review management automation, AI-assisted CRM workflows, proposal generation tools.

The tools are not identical. But they’re close enough that buyers can’t tell the difference — and increasingly, price is the only lever left.

David Long — 25 years in software development, building a voice agent for missed calls — said it plainly: “After I was building it, I found a lot of competition, and the price point kept getting lower for a similar product.”

This is what commoditization feels like from the inside. You’re not imagining it. It’s happening.

The Commoditization of Intelligence

The underlying models — GPT, Claude, Gemini, ElevenLabs for voice — are becoming infrastructure, not products. In the same way that nobody competes on “we use AWS” anymore, competing on “we use AI” is rapidly becoming meaningless.

The specific capabilities that seemed like advantages a year ago — natural language understanding, multi-turn conversation, tone matching, document summarization — are all available to any developer in an afternoon. The moat evaporates the moment the capability becomes an API call.

When Ian was asked whether David should keep building his voice AI or pivot, the answer cut to the core: “AI itself is becoming the commodity. The know-how lies in the mid-layer.” That mid-layer is not a feature. It’s a body of knowledge.

Why Workflow Knowledge Matters More Than Models

Here’s what AI cannot do on its own: understand your client’s business.

It doesn’t know that the sales process at a mid-size furniture importer runs through a single WhatsApp thread. It doesn’t know that the qualification criteria for a luxury watch buyer looks nothing like the criteria for a medical device buyer. That knowledge has to come from somewhere — and wherever it comes from, that is the actual product.

Ian’s approach at ADAIA makes this explicit. When building a sales assistant for a client, the process is not “pick a model and deploy it.”

The result is an agent that behaves like a trained team member in that specific business — not a generic assistant. That specificity is the product. And it’s not reproducible at scale by a platform.

How Consultants Can Outperform AI Startups

Swetank Jackson — a SaaS engineer in Vienna selling AI systems door-to-door to local businesses — made an observation that landed perfectly: “Building it reversely — not building what you want, but building what the pain point is, and then doing the engineering from there.”

That’s the structural advantage consultants have over product companies. A product company has to generalize. A consultant goes deep into one context. They learn the workflows, the edge cases, why the previous CRM failed. They accumulate knowledge that a product cannot ship in a box.

“They couldn’t care less about what system you’re using under the bonnet. They just need the results — and they can’t get the results on their own because they are not the experts who can transform their business knowledge and processes and policies into system instructions.”

That transformation — from business knowledge to AI-executable instructions — is the work. And right now, it’s almost entirely a human job.

The Future Belongs to Vertical AI

The one category of AI product that remains defensible: vertical AI built around deep, accumulated domain context.

Not “voice AI.” Voice AI for Austrian dialect recognition, specifically targeting DACH businesses where cold contact regulations require in-person permission-first outreach. That product has a different customer, a different regulatory constraint, a different technical requirement, and a different acquisition path.

Ian’s advice for building in a narrow vertical: “I would do a little bit of the research on the technology itself, maybe a small short demo, and then I would go straight to the guys who would say, ‘Yeah, we’d acquire you if you show traction.’”

Build it already aimed at the exit. Know who your buyer is before you write the first line of code.

Quick Tips: Building Around Context

• Pick one niche and go deep. Depth of industry knowledge compounds. Width does not.
• Learn BPMN. It is the most underused tool in AI implementation. A workflow diagram is worth more than an advanced prompt.
• Treat your domain knowledge as IP. Document edge cases, escalation rules, client-specific policies. That documentation becomes your system instructions — and your moat.
• White-label the commodity layer. Use ElevenLabs, VAPI, Make.com. Don’t build what’s already been built. Build the layer that sits on top.
• Your clients don’t care what’s under the bonnet. They care about results. Your job is the context translation, not the model hosting.

Frequently Asked Questions