Here’s the problem: when you build AI solutions on a platform-by-platform basis, you’re creating silos and proving Conway’s Law in the AI era. Siloed teams using siloed platforms create siloed applications, and that inefficiency scales at the speed of automation.
Simply put, when your teams aren’t communicating effectively, you end up building solutions that satisfy the requirements of one team but not the company as a whole.
What you end up with is RevOps building agents in Salesforce, Finance building them in NetSuite, and Engineering building them in their own stack, with nobody able to communicate with each other or reuse any components.
Think about what should be possible: RevOps publishes tools that Engineering can pull from to show subscription and invoicing information in-platform. Finance creates data models that Sales can leverage for forecasting. Operations builds workflow components that every department can plug into their processes.
This isn’t science fiction; it’s basic modular architecture applied to AI tooling. But it only works if everyone’s working from the same sheet of music.
The Platform Principle for AI: Having teams on the same platform allows you to have a cohesive, unified, understandable framework that teams can use to speak to each other. This is what reduces bottlenecks and drives efficiency. Not AI itself, because AI without people and process improvements and management just repeats the same mistakes we’ve made over the last 20 years with SaaS: just at a greater and faster scale.
The Vendor Landscape Is Fighting Yesterday’s War
Currently, we’re witnessing the gap between AI hype and reality play out across vendors. Some are approaching this with measured realism, preparing their boards for a different future in earnings, willing to jettison products when it makes sense. Others are doubling down on walled gardens, trying to make their slice of the pie bigger rather than expanding the pie itself.
This is blue ocean strategy at play - create an uncontested market space rather than competing in existing markets.
The winners in this environment realize that when you’re in massive change, you don’t quibble with competitors—you focus on strategy and product. You don’t hold onto sacred cows that must remain in place.
Salesforce’s Walled Garden Problem
This is why I think Salesforce’s AI strategy is ultimately flawed. We all know Salesforce has wanted to be a data company for the last two decades. After all, it’s why we have the phrase “if it’s not in Salesforce, it doesn’t exist.” It’s why we have so many vendors push data into Salesforce: call recording, notes, everything. It’s why we saw the experiments with data.com and later the acquisitions of MuleSoft and Informatica.
But think about the fundamental absurdity here: why are we storing files in Salesforce when it’s the most expensive storage rather than S3 or OneDrive or Google Drive? Why are we not using the best tools for the job that can be used ubiquitously across the enterprise?
Salesforce has pushed Data Cloud with mild adoption rates when we’ve been shoving data into our data warehouses for years—the same warehouses that already have our data, have pipelines between our SaaS sources (HRIS, ERP, CRM, OMS) plus our core domain product data if you’re a product company.
We see Agentforce’s abilities grow and mature, but it only works inside Salesforce. What if we want to build agents in Finance? In Operations?
The Salesforce AI strategy today:
Charging customers to ingest data they already have in their warehouses
Building agentic flows that don’t extend outward easily
No application layer to build better apps
Customers are still complaining about the UI with no alternatives
As SaaS platforms start to decrease in importance and become more of the database layer, how do these walled garden strategies play out? We see Salesforce’s CEO, Marc Benioff, constantly belittling other companies' strategies while ignoring that his community is turning against him. He quibbles with ServiceNow and tries to create a service business. We see agentic products bolted onto Salesforce offerings.
Here’s what’s wild: Salesforce has the most unique positioning with a third-party connector, MuleSoft Composer. You’d think they could use this positioning to make Composer the agentic competitor to Zapier and Workato, allowing them to diversify their strategy. They’re investing heavily in Slack, but once again, this feels like a disconnected experience when the Salesforce umbrella has so many products underneath it.
The Microsoft and ServiceNow Gambit
Contrast this with Microsoft. While they stumbled out of the gate with Copilot in the enterprise, they’re doing a decent job with Power Apps and their GitHub products. They’ve also signaled that they’re jettisoning their CRM products.
ServiceNow is entering the CRM space, which is interesting. Why would they invest the time and energy to enter this crowded market (Oracle, Salesforce, HubSpot, Microsoft)?
All this indicates that companies are realizing they need to own everything in order to stay relevant, so they’re straying from their core competencies. But they’re all sluggish.
The weakness with Microsoft is that it relies heavily on its partnership with OpenAI rather than building its own technology like Google. OpenAI pulling the rug from under them as they enter into competing territory to make money could be devastating, and I think Satya Nadella knows it.
Neither Microsoft’s (Copilot) nor Google’s (Gemini) productivity AI tooling is wowing users, but I think that will change in the coming years. It’ll be interesting to see how Google approaches the next steps with AppSheet as their answer to Power Apps, and it’s not very good.
The Third-Party Orchestration Layer Opportunity
CTOs and CFOs don’t want duplicative spend. Whoever can solve this wins. CFOs don’t want to buy cutesy names in 2025; they want to buy value.
We saw companies advertise to business teams in Marketing and Sales over the last 15 years. The new focus will be on cost-conscious finance teams that are facing continued pressure from their boards to grow margins, especially in a cash crunched capital market.
This is where third-party AI orchestration layers stand poised and ready to connect applications and experiences. Platforms like n8n, Make, Workato, Zapier, and Retool are completely agnostic of the platforms they interact with and can adapt as individual apps are replaced.
What this enables:
Buy once, cry once; reducing overall maintenance costs
Increased ramp speed as the company works across the same sheet of music
Stitch processes together while using the AI capabilities of their own platforms
Reusable components across departments and teams
SMBs and mid-market companies have an opportunity to be more agile here. Using AI-centric ERPs like Rillet, combined with a multitude of CRMs or service platforms. They can utilize AI-centric dialers and note-taking apps, leveraging AI orchestration layers to integrate these applications.
However, we also see Microsoft poised to play a major role itself. They’ve spent decades in the enterprise space, and while they don’t have the sexiest products, they do become easier to use with AI. Power Apps connects to a wealth of information across OneDrive, Email, Teams, and SharePoint. While this isn’t sexy to the average SMB, this is immensely powerful in full Microsoft shops, especially when you layer in access to Azure and core infrastructure often running in these orgs.
The key question: Will we see third-party startups that are agnostic ramp up because of agility? Or are they priced out? That remains to be seen.
What You Should Actually Do
As you evaluate your AI platform in your organization, it’s important to realize that you can experiment now with various platforms at a small scale. But it’s far more important to start organizing your data in such a way that you can use it in any agentic platform.
Your AI readiness checklist:
Start documenting all the institutional knowledge that’s in folks’ heads (your AI strategy cannot tap into undocumented knowledge)
Get your employees comfortable with AI strategies and techniques
Provide stepping stones, not ultimatums
Organize your data to be platform-agnostic
Choose one or two enterprise-wide agentic platforms, not one per SaaS tool
Don’t thrust AI upon your teams. Provide stepping stones to help them become comfortable. Because then they’ll start to understand it better and how it can apply to them every single day. Encourage your teams not to generate slop but to use their brains and let AI augment it.
If you thrust your employees into AI-first ultimatums, this is what you get: slop to prove that they’re indeed using AI. That’s not delivering any value.
The most important thing: Have a cohesive strategy about how you use AI across the enterprise. We firmly believe that it should be one or two agentic platforms and not one on each SaaS platform. Having teams on the same platform allows you to have a cohesive, unified, understandable framework that teams can use to speak to each other.
Measure What Matters
We shouldn’t focus on the number of agents we have in production, but rather the quality of them and what value they’re delivering based on measurable results.
Focusing on the number of agents is like focusing on the vanity metrics of the 2000s. We focused on website visitors and session length, not what really mattered: conversion. You can have a million visitors, and if none convert, it’s worthless. Same thing with agents. You can have thousands running, but if you can’t tell their impact, you’re just spending money and time building and maintaining them, likely just so you can tell your boss or your board that you’ve indeed rolled out AI.
Smart players in the agentic future will:
Focus on expanding the pie, not fighting over budget and territory
Work with teammates and other departments collaboratively
Avoid the “Shadow IT” sprawl that plagued 2010-era SaaS adoption
Build extensible and modular platforms, not bolt-on solutions
Because in many ways, RevOps is seeing “Shadow IT” for the first time, the same way that IT saw it on the rise in 2010. The stepping around teams caused massive sprawl and inefficiencies then, and it’s going to cause it now.
Think Like a Product Engineering Team
Thinking about AI enterprise-wide allows you to be nimble and agile, adapting with the next waves in technology that will happen. If one SaaS platform folds, you’re okay. You know that you can rebuild around another area (assuming that your platform of choice doesn’t fold).
However, if you’re locked in with each vendor’s agentic platform, you have to rebuild each time. That’s real time and money lost. Worse, you’re at the mercy of each vendor’s AI strategy. If one lags, then you lag behind your competitors.
The extensible architecture approach:
Stop cramming all of your data into one SaaS vendor like you have in the past. The irony of vibe coding is that most of the vibes are spent building inside the box rather than outside of it. Think like a product engineer and realize that we diversify our tech for a reason. We choose the right tool for the right job.
Storage like S3 is amazing and allows you to connect your files to all tools (and ingest into your vector database)
Databases like Postgres allow you to render information that any of your tools can use, and display that information elsewhere (this is agentic-friendly!)
Use vibe coding to extend your platforms, not replace strategic thinking
Learn patterns like SQS/SNS or Redis caching to save money on AI builds and decrease latency
You wouldn’t vibe your company’s selling strategy. Don’t vibe how you run your business.
The Window Is Closing
Here’s what most executives miss: the decisions you make about AI architecture in the next 12-18 months will determine whether you’re nimble or paralyzed for the next decade. This doesn’t mean that you have to rush into execute, but it’s time to begin experimenting and figuring out what is right for your comany.
Every platform-specific agent you build today is technical debt tomorrow. Every department that spins up their own AI solution is another silo you’ll have to untangle later. Every vendor lock-in you accept is leverage you’re handing over in future contract negotiations.
The companies winning in three years won’t be the ones with the most agents in production. They’ll be the ones who built extensible, modular, enterprise-wide platforms that let them pivot when the technology shifts; and it will shift.
They’ll be the ones whose RevOps team can publish a tool that Engineering reuses the next day. Whose Finance models feed directly into Sales forecasting without custom integration work. Whose cross-functional teams speak the same language because they’re working on the same platform.
Your competitors are making their architecture decisions right now. Some are building on quicksand with vendor-specific solutions. Others are thinking holistically, bringing their people along for the ride, and realizing how powerful unified platforms can be to tear down barriers.
Which camp are you in?
Want help navigating this transition? I work with CFOs, CTOs, CIOs, and RevOps leaders to design enterprise-wide AI strategies that avoid vendor lock-in and actually deliver measurable value. Let’s talk about your architecture.
Subscribe to get insights like this in your inbox and stay ahead of the AI platform wars.