Is Salesforce Dead?

The last tech-related post I posted on this blog was after being at the Agentforce World Tour in London back in December. At the time, Salesforce had just announced Agentforce 360 and the entire event felt like a huge statement about where enterprise AI was heading. The messaging was everywhere: AI agents, autonomous workflows, agentvibes... Companies like Tiffany and LIV Golf were already being showcased as examples of the future.

Fast forward a few months and things suddenly feel very different.

Salesforce stock has taken a 25% hit despite still showing year-over-year growth. Every week I hear someone say they have “built their own CRM with AI” or that SaaS companies are dead. And funny enough, LIV Golf has had funding cut from the PIF….

And honestly, as someone who is always playing around with MVPs and dummy tools, I understand why people are starting to believe it, models like Opus 4.6 are mind blowing…

That said, as someone who build and maintains a Salesforce instance in a company belonging to a conglomerate with 100K+ employees (and a lot of IT rules), I can confidently tell you that Salesforce is not dying anytime soon.

What is happening is far more interesting: 

AI is changing what CRM systems are actually for.

A lot of the current discussion around Salesforce treats it like it is just another SaaS platform with expensive licenses and a clunky interface.But at enterprise scale, Salesforce is rarely just “a CRM”.

If implemented properly, it becomes the operational memory, the “brain” of a company. It knows your customers, your pipeline, your sales cycles, your support history, your products, your relationships, your approvals, your contracts and increasingly the context behind every interaction. There are very few systems inside a company that understand the business better than a well-designed CRM.

Whether you are B2B or B2C, a well built CRM often knows your customers better than your best sales rep does. And that is incredibly important in the age of AI, because AI without trusted business data is just a very confident guessing machine.

At the same time, the “SAAS haters” are not wrong. AI has fundamentally changed software development.

Today, a startup can absolutely vibecode a CRM instead of paying tens or hundreds of thousands every year for a traditional enterprise platform. If you are a small company with limited funding, I actually think you should 100% rely on Claude to build your first CRM.

The barrier to building software has collapsed. But the barrier to maintaining enterprise-grade software has not, if anything it has gotten harder than ever, and that is the part people massively underestimate.

I say this from experience: eventually many internally-built systems become monsters. At first, AI makes development feel almost magical. Features appear incredibly quickly. Problems get solved in minutes instead of weeks. But AI is so good at producing code that the amount of code grows faster even than most teams can properly understand and document.

Every week there are new dependencies, new vulnerabilities, new frameworks, new integrations and new maintenance requirements. Suddenly the thing that saved you money starts consuming all of your time just to stay operational and not become a liability. 

At enterprise scale, paying another company a huge amount of money to manage infrastructure starts making much more sense than people think. You are not just paying for screens and objects.

You are paying for:

Security, governance, uptime, scalability, compliance, data residency, integrations, support, releases, patches, legal protection, and most importantly, you are paying for trust.

Pivot or disappear

That said, I do think Salesforce will have to adapt to survive, it’s still not getting it's clients ready for the agentic world.

The biggest constraint in almost every CRM project I have worked on has never been the technology itself, It has always been the users.

Before releasing any new functionality, I usually think about four things:

1. How much are users going to hate this new field or validation rule?

2. How much value does this actually add?

3. How much can users accidentally mess up this input?

4. How do we make the experience as simple as possible?

Because the reality is that most people do not enjoy updating CRMs. There is nothing more frustrating than finishing a long day and suddenly remembering you still need to update Salesforce. Even worse when a validation rule blocks you at 5 pm on a friday because you forgot to enter the industry for a company everyone already knows.

For years, CRM architects have been trying to solve this problem through better page layouts, conditional visibility, automation, enrichment tools and UI improvements. But fundamentally, we were still asking humans to manually structure data all day long and we all know how error prone that tends to be.

And that is exactly where AI changes everything.

AI Is Turning CRMs Into Infrastructure

Over the last year, I have watched processes that used to require entire enrichment platforms and weeks of thinking to become a single LLM call.

Matching duplicate accounts, extracting information from emails, summarising meetings, enriching company records, identifying sentiment, categorising opportunities, things that previously required complicated and quite useless NLP pipelines can now happen instantly.

And in the last few months, with the rise of MCPs and agent-based systems, CRMs are starting to move away from being the front-end employees constantly interact with.

Instead, they are becoming the trusted knowledge layer underneath AI systems.

This is the shift I think many people are missing. The future CRM is probably not something users spend all day clicking through. The future CRM is the system AI agents read from and write to on behalf of users.

Instead of manually updating opportunities, sales reps will approve AI-generated updates from meeting notes. Instead of searching through consuding reports, users will ask questions in natural language. Instead of forcing users to fill out endless forms, AI will continuously enrich records in the background.

In some companies this future is already starting to happen. Headless CRM systems, conversational interfaces and AI-assisted workflows are turning Salesforce into something closer to enterprise infrastructure than enterprise software.

Ironically, AI may actually make CRM systems more important, not less. Because the better the AI becomes, the more valuable trusted and structured enterprise data becomes.

The Pricing Problem

This is where I think Salesforce genuinely has a challenge… The traditional per-seat pricing model starts breaking down in a world where AI agents perform work on behalf of thousands of employees.

Hypothetically, an entire Salesforce instance could eventually be maintained by a handful of user licenses giving access to fleets of AI agents doing the operational work, and that completely changes the economics of enterprise software.

And to Salesforce’s credit, they clearly understand this. The company is already moving toward usage-based and consumption-based AI pricing models.

I just do not think the transition is happening fast enough yet. I don’t think companies are ready to fully trust black-box AI systems anytime soon. The moment users spot one incorrect AI-generated update, confidence collapses incredibly quickly. Trust in enterprise data is

fragile and once damaged it is extremely difficult to rebuild.

That is why I think the near future will be hybrid. AI will do the heavy, boring lifting: summarising calls, extracting actions, enriching records, suggesting updates, identifying missing information…

But humans will remain in the loop and will be kept accountable for the results. I imagine AI as an extremely brilliant intern, that for good or for bad, his manager will take the credit or the blame for its performance.

I truly believe accountability drives trust, and nothing drives adoption more than trust.

With that, I wanted to finish off with Three Things I Would Recommend To Anyone Using AI for/with a CRM

1. Fix the pipe before you clean the pond

Focus on the processes generating bad data before trying to clean historical records. Otherwise you end up endlessly correcting the same problems over and over again.

2. Focus on data before AI

If users realise your AI is operating on unreliable data, trust disappears immediately. Good AI on bad data is still bad, once you have the right data it will be easy to add an AI system on top.

3. Use AI to create better data for better AI

This feedback loop is very real. AI is incredibly good at enriching metadata, categorising information and improving record quality. Better metadata then improves future AI outputs, which improves the data even further. And importantly, you do not always need the most advanced models to do this effectively. A lot of enrichment tasks can be done cheaply in batches using smaller or lower-cost models.

Final Thoughts

I think AI will absolutely take a significant bite out of Salesforce’s current business model. But I also think it positions Salesforce to become something even more valuable: the trusted enterprise data layer that AI systems depend on.

The companies that win in the next decade will not necessarily be the ones with the flashiest AI demos. They will be the ones with the most trusted data. And despite all the noise around SaaS being dead, I still think Salesforce is one of the best-positioned companies in the world to benefit from that shift.

Although Salesforce seems to not be going anywhere anytime soon, I am not so sure about LIV Golf…