Build vs Buy AI: When Your Custom ChatGPT Platform Becomes an Achilles Heel

Ethan Mollick, Associate Professor at Wharton and author of Co-Intelligence, recently shared a striking observation on LinkedIn:

"A sign of a company being advanced in AI adoption was that they built their own internal chatbot (usually called 'CompanyNameGPT') using APIs.

As the major lab's chatbots become agentic, bringing together many tools in a single interface & as they add memory & projects, the custom API chatbots are falling behind.

They may become a liability that holds back the companies that were fast movers."

If we spoke recently and you're reading this and thinking I'm talking about your company, you're right. But you're also not alone - I have this conversation repeatedly with enterprise clients who have built internal tools.

Why Building Made Sense 18-24 Months Ago

Building a custom AI platform was not a bad decision in 2022 or early 2023. There were legitimate reasons IT leaders chose to build rather than buy:

Security and governance concerns were real. Enterprise vendors didn't have ISO 27001 certification, SOC 2 compliance, or the data residency guarantees that regulated industries required. Building internally meant controlling where data lived and who could access it.

Vendor integrations didn't exist. OpenAI's ChatGPT Enterprise didn't connect seamlessly to Microsoft 365, Google Workspace, Hubspot, or Box. If you wanted AI to access company knowledge bases, you had to build those connectors yourself.

Multi-model flexibility was attractive. IT teams saw value in giving employees access to models from multiple vendors - OpenAI, Anthropic, Google - in a single interface rather than being locked into one provider's ecosystem.

These were sound strategic reasons. Your team wasn't wrong to build.

The Landscape Has Changed (Fast)

Here's what's different today:

Governance is no longer a differentiator. ChatGPT Enterprise now has ISO 27001, SOC 2, GDPR compliance, and data processing agreements that satisfy most legal and compliance teams. Vendors specifically exclude any model traing on Business and Enterprise plans. The security gaps that justified custom builds have largely closed.

Vendor ecosystems are maturing rapidly. ChatGPT connects natively to Google Drive, Microsoft OneDrive, and can pull from internal knowledge bases through enterprise search integrations. Anthropic's Claude offers similar enterprise connectors. The integration work that required custom development is now available out of the box.

Agentic capabilities are accelerating. Memory, projects, voice, images, multi-step workflows, and tool use are shipping monthly from major vendors. Internal teams are now competing with well-funded AI labs in a features race they can't win.

Your employees already have ChatGPT on their phones. The consumer experience sets the bar. If your internal platform is slower, clunkier, or missing features employees see in the public app, they'll route around it.

The IT teams who were "fast movers" 18 months ago are now in a different race: keeping up with vendor development velocity.

The Hidden Cost: A Story About Email

Let me tell you about a major European discount supermarket chain I worked with around 2018.

During our engagement, we struggled to schedule meetings. Calendar invites bounced. Emails arrived with strange formatting that broke in Outlook and Gmail. Simple logistics became friction points.

When I asked my contact why communication was so difficult, he gave a wry smile. "The chairman's son is a very keen technologist," he explained. "He built our own in-house email and calendar system."

The problem? It didn't interoperate well with Microsoft or Google ecosystems. Internal employees could message each other, but communicating with anyone outside the organization - customers, vendors, partners - was painful.

No one was brave enough to raise this with senior leadership.

I don't know if that company ever migrated onto Google or Microsoft. But I think about that email system whenever I see companies defending custom AI platforms that duplicate commodity functionality.

Jeff Bezos has a useful quote here: "Focus on what makes your beer taste great." Email isn't a competitive differentiator for a supermarket chain. And for most companies, the user interface employees use to access AI isn't what makes your business unique.

The Build vs Buy Audit

If your company has a custom AI platform, here are three questions to ask:

1. What Does Your Roadmap Look Like Against Vendor Capabilities?

List the features shipping in the next 6 months from your internal team. Now compare that to what OpenAI, Anthropic, Google, and Microsoft shipped in the last 90 days.

Are you keeping pace? Falling behind? Duplicating work that vendors will release as standard features? Is this what your IT team should be working in?

2. What Do Employees Actually Think?

Interview 10-15 employees who use your platform daily. Ask:

• What works well?
• What's frustrating?
• Have you tried ChatGPT, Claude, or other consumer tools? How do they compare?
• If you could change one thing, what would it be?

You might be surprised by the gap between IT's perception of the platform and user experience.

3. What's Changed Since You Built This?

Three years ago, ChatGPT Enterprise didn't exist. ISO 27001 certification and data residency wasn't available. Data processing agreements didn't meet enterprise standards. Native integrations to company systems weren't an option.

Today, those objections are moot points. What reasons remain for building internally versus buying from a leading vendor?

When Building Still Makes Sense

There are legitimate cases where custom development remains the right choice:

Highly specialized industry workflows. If you're in pharma, financial services, or defense with workflows that require deep customization beyond what vendor APIs offer, building might still be justified.

Proprietary data moats. If your competitive advantage comes from unique data sets and models trained on that data, building infrastructure to leverage that IP makes sense.

Mission-critical integrations. If vendor platforms genuinely can't connect to your legacy systems and those integrations are business-critical, custom work may be required.

But for most companies and most roles within those companies? The user interface for accessing AI is not your differentiator. Your product, your customer relationships, your domain expertise - that's what makes your beer taste great.

What to Do Next

This article isn't about making you feel bad for building something 18-24 months ago. It's about creating space to ask: Is this still the right strategy?

The fast movers who built first now have an advantage - you understand AI's impact on your business. You've identified use cases. You have engaged users.

The question is whether continuing to invest in custom platform development is the best use of that momentum, or whether partnering with a leading vendor lets you focus resources on what actually differentiates your business.

Is Your AI Strategy Keeping Pace?

If this article resonates with your situation, you're not alone. We work with enterprise IT and business leaders navigating exactly this transition - from custom builds to strategic vendor partnerships that free up resources for real differentiation.

Start chatting for free to explore how leading vendors' platforms stack up against your current capabilities, or see how Expert Requests work to get strategic guidance on build vs buy decisions.

Go Deeper

Want more practical guidance like this? Subscribe to Kowalah Insights.

• OpenAI ChatGPT-5.1 release blog post
• Kowalah Wednesday Webinars - learn one ChatGPT feature and ask your questions
• Kowalah Templates and Guides - AI policies, job descriptions and more - all free and un-gated