Recently, I spoke with a colleague who spent a weekend building a new analytics site with an AI coding agent. His team gave it rave reviews, saying that they liked it better than the product it replaced.
The day we spoke, however, that colleague was also dealing with an outage. His site wasn’t loading, and the team had gone back to the original product to get their numbers. The reason? The new site pulled every number from the product it was meant to replace, and the vendor closed the security loophole.
I’ve heard the same story a few times, though they all sound different on the surface. A founder expects an AI model to run a game’s core logic on its own.
Executives who tried an AI coding tool at home are asking their CIOs to rebuild systems the company already owns, ERP included. In each case, the builder treats the AI’s output as the business.
AI is a tool, not the business. AI may be able to rebuild your software this weekend, but the durable value your business has created takes far longer to build.
I call the force behind these weekend projects the Great Compression, and it is making its way into every product line and internal system we run.
The Great Compression
Software that once took a team a year now takes a capable person a few weekends, and buying decisions already reflect it.
In McKinsey’s 2026 global survey of 1,719 respondents, fielded in May and June, 32% said their organizations had decided against buying at least one software product or feature because they could build it internally with agentic coding tools, and in technology companies the share was 41%.
A large language model, and every agent built on one, compresses much of what people have written into a probabilistic map and generates the likely next step along that map, and the training that makes models helpful narrows that range further.
Researchers have measured the result, finding that different models give strikingly similar answers to the same open-ended questions, and model-written code solves problems with fewer distinct approaches than human-written code.
Weekend rebuilds of the same product therefore tend to look alike. They copy the parts of a product you can see in their most common form, leaving behind much of what made the original distinct.
Consequently, the question for every leadership team is simple. When someone can rebuild our software in a weekend, what is left that we own?
The Weekend Test
I answer that question with one test. For each product and major system, you should ask what a weekend rebuild of it would still depend on.
“Whatever the rebuild depends on is your moat, and everything it can copy is a cost to manage.“
The analytics site shows how the test works. The rebuild copied the screens, but it still depended on four things that provided enduring value. Those were the customers who trusted the original product, the years of data behind every number, the company standing behind the service and the partners who keep it running.
Your Customer Base
A few years ago I helped build a security startup whose product predicted attacks for its customers. Our model ran inside each customer’s own firewall, next to data the customer would never have sent to an outside company.
A well-funded competitor could have copied the model. However, it could never have copied the trust that got us invited inside the firewall in the first place, earned over months of proof and a track record our customers could verify.
Large buyers still act on that kind of trust. In a January 2026 survey of executives at 100 Global 2000 companies, 65% said they prefer incumbent AI solutions when one is available, citing trust, integration with existing systems, and simpler procurement.
A weekend rebuild starts with no customers and no track record.
Your Accumulated Operating Data
Your data is the years of transactions, lessons, exceptions, and fixes that chronicle how your business actually runs.
Every product I’ve built with AI has taught me the same lesson. The code came together fast, but the lasting value came from how the data fit together, something no model could write in a weekend.
Most organizations can’t use that history yet. In a 2026 survey of executives at 2,000 companies, Accenture found that 72% lack trusted, well-governed data of the quality advanced AI needs, and only 7% have data ready to scale it.
Until your teams and your agents can use the history, the history is a cost you pay to store.
Data also has a limit. Chegg spent years building a large library of solved homework problems, and its revenue fell 48% from a year earlier as Google’s AI answers and students’ own AI tools drew students away.
The library still exists. Accumulated data survives only while customers still need what it produces.
Your Finances
When anything can be built in a weekend, customers buy from vendors who will be there to stand behind it.
In 2023 Microsoft told its commercial Copilot customers that it would defend them against copyright claims over Copilot’s output and pay adverse judgments, subject to conditions.
A weekend project can write the same promise, and no customer will believe it.
Cash also buys time and customers. In 2025 Salesforce agreed to buy Informatica for about $8 billion to supply data to its AI agents, and it completed the purchase that November.
The deal bought years of data integration work and the customers who depend on it.
CIOs influence this survivor more than most of us admit. Money we stop spending on what AI can copy can go to what it can’t.
Your Partnerships and Relationships
A competitor can match your price list tomorrow. A competitor can’t match a partner that has spent five years in your annual planning, embedded engineers within your teams, and sent its own people in when a critical system fails.
The trust behind a strategic partnership takes years to earn on both sides.
This survivor belongs to the CIO more than the others do. You can delegate day-to-day vendor management, and you should.
However, you should own the strategic relationships personally, share your plans with those partners, and build arrangements where both sides win.
Your Job on Monday
It’s our job as CIOs to find our moats and protect them. Run the Weekend Test on your top ten systems and your top ten vendor contracts and stop spending scarce engineering time on anything a weekend rebuild could copy.
Give every weekend build an owner. The analytics site that went down and the game built in a weekend still need someone accountable for fixing them, securing them, and paying to run them.
Each of the four moats needs an owner too. Someone needs to be accountable for customer trust, operating data, room in the budget to invest, and strategic partnerships.
Weekend projects are useful signals. They show where people need help and where the products we buy fall short, and now anyone in the company can build one.
The CIO’s job is to turn the good ones into supported tools and keep building the customer trust, data, finances, and partnerships the business runs on. That CIO will move the whole organization faster with AI, and will have the business results to show for it.
