Opinion
What Salesforce taught me that Israeli tech never had to learn
According to Tribal CEO Yoav Kolodner, the software you sell is a third of what the customer actually runs. The rest is enterprise context, it is where 95% of AI pilots die, and it is the part Israeli founders almost never get to see.
I spent years as VP of Engineering at Salesforce. Again and again I'd sit with a customer who couldn't say what a field was for, or which automation had created it, inside a system they'd been running for a decade.
Nobody could say what would break if they removed it either.
Two of them would buy the same licence, the same release, the same everything we shipped, and five years later have nothing in common. One had fifteen years of custom objects, validation rules, approval chains and integrations stacked on top of the platform, built by people who had left the company, documented nowhere. The other had built something entirely different on the same foundation. Two different pieces of software, and neither of them was ours.
What stayed with me was watching how that played out for the teams inside those companies. They stalled constantly. Not for lack of engineers or ideas, but because every change ran into something nobody could see. A governance rule nobody could locate. A dependency nobody had mapped. A documentation gap where the person who wrote the automation left in 2019. The work was not hard to do. It was hard to establish that it was safe to do.
That is what enterprise software actually is, and it is not what it looks like from outside. The product is maybe a third of it. The rest is a decade of undocumented decisions that nobody in the building can fully explain and nobody can safely remove.
What those teams were missing has a name. Enterprise context: what exists in the system, what depends on what, and what breaks if you touch it. Nobody had it, so nobody could move.
I think about that constantly now, because it is exactly what enterprise AI keeps walking into.
Related articles:
- Former Salesforce, Wix and Spot.io executives raise $10 million for enterprise AI agent platform Tribal
- Salesforce veterans raise $20 million pre-Seed from Marc Benioff and Michael Dell to tackle AI’s enterprise bottleneck
- Salesforce Ventures invests an estimated tens of millions in Upwind at $1.6 billion valuation
MIT's NANDA initiative found in July 2025 that roughly 95% of generative AI pilots produced no measurable impact on the P&L. Gartner predicted in June 2025 that more than 40% of agentic AI projects would be cancelled by the end of 2027, citing escalating costs, unclear business value and inadequate controls. Model quality appears on neither list.
The failures share a shape, and it is the same shape as those stalled teams. The pilot works, because it runs inside a clean system or a sandbox or a well-behaved corner of one platform. Then it meets the real environment and produces something that looks correct and that no IT department will approve, because it cannot tell which automation fires on which table, or who has permission to change what, or what breaks two systems downstream.
The model is not the problem. It is generating confidently into an environment it cannot see, which is what those engineering teams were doing before they learned not to.
Most Israeli founders have never seen this up close, and I do not say that as criticism. It is a function of where we build. Cyber, adtech, developer tooling, all fields where you start from nothing and the customer's accumulated mess belongs to somebody else. The maintenance economy around enterprise software sits in India, in Europe and in the American consultancies. We sell into enterprises. We have not had to live inside one.
That is worth changing, because there is an enormous market sitting in the gap I spent years staring at. Ninety-five percent of current enterprise AI spend is producing nothing, and the reason is not intelligence. Somebody is going to solve that, and solving it means work that demos badly. Mapping what a customer already built. Establishing what a change will touch before making it. Treating the environment as the product surface rather than as an implementation detail.
None of that wins a keynote. Israeli tech has historically been very good at unglamorous technical problems once it decides they are worth solving.
What I took from Salesforce is that the software is the easy part. What sits on top of it is the business, and until AI can see that, it will keep failing at exactly the point where it starts to matter.
Yoav Kolodner is co-founder and CEO of Tribal.
