The AI myths holding revenue teams back

Engineering teams using Claude Code and Cursor are at least ten times more productive than they were two years ago. One engineer now does the work of ten.

At Von, 95% of the code is written by AI. Ten people are doing what used to require a hundred.

If you told one of those engineers you were taking Claude Code away, they would leave the same day. Now ask the same question about your sellers, your managers, and your CROs. Is there a single tool you could remove that would make them walk? Salesforce? You would probably hire more people if Salesforce disappeared.

That disparity, between what AI does for engineering and what it does for revenue teams, is being protected by five myths.

Five years ago, I started a company called Rattle. Rattle is a workflow automation platform, which you'd call typical legacy SaaS. We have more than 200 customers, and these are companies anyone would dream of working with. Half the billboards you see on the 101 are Rattle customers. Figma, Intercom, Writer, and many more.

Now, we're pivoting the entire company into Von.

Von is only a few months old and already has 45 customers and building an AI-native product has shown me exactly why Salesforce is down 40% in the last 12 months and why HubSpot is down 70% in the same period.

The value AI products deliver is so disproportionate that you won't pay for legacy SaaS the same way again. At least not at the same price. If you can pay $200 a month to make your engineer 10 times more productive, your engineering team will stop paying $200 a month per seat for a SaaS platform that makes them 10% more productive. That's the math the market has woken up to. There's real panic out there. I'd short SaaS companies if I were a betting man, which I'm not.

We spent $3.5 million on R&D over the last 12 months to get Von right, so let's get into what we learned along the way.

Myth 1: My CRM is too messy for AI to work

We have been talking about data hygiene for 25 years. If you attended a Salesforce launch event in the early 2000s, someone was on stage saying the same thing. The data is never going to be fully clean, and waiting for it is no longer a viable strategy.

The myth misunderstands where bad data actually originates. Your raw data, sitting in call transcripts, emails, and seller notes, is not the problem. The problem appears in translation. When sellers move that raw data into Salesforce, errors enter. You then run reports on the degraded version and wonder why the insights feel wrong.

Connect AI to the raw data and skip the polluted version entirely. Von normalizes the last two years of data by going directly into call transcripts, emails, and seller notes. It determines what the deal amount should have been, who the competitor was, what the close date should have been, and when each deal should have moved stages. Once that normalization is done, you already have what AI needs.

💡

One prerequisite: your calls need to be recorded.

If European regulations make that impossible, tools like Granola transcribe locally without joining as a bot or recording video. If even that is off the table, high-level seller notes combined with email threads are a workable foundation.

Myth 2: If we just tell sellers what to do, they'll do it

The recommendation model was borrowed from consumer platforms. YouTube, TikTok, and Instagram surface recommendations that people engage with constantly, so the assumption was that the same logic would hold in B2B sales.

It does not. Those platforms run on extraordinary volumes of behavioral data. YouTube alone accounts for roughly 2% of all human time globally, according to Alphabet's 2023 shareholder materials. That is the depth of signal feeding those engines.

In B2B, you have neither the data volume nor the compute capacity to build anything equivalent. So you run the same generic playbook across every seller and every CSM, and it fails because seller A prioritizes accounts where the contact has recently changed jobs, while seller B focuses on companies that have just migrated to a new platform. They are working from different mental models of what a good opportunity looks like.

What actually works is asking your top AEs and CSMs how they rank deals and what signals matter to them. When a seller can test a hypothesis against live data in minutes, without routing a request through BI or RevOps, the dynamic changes entirely.

A seller says, "I think customers churn for these three reasons. Validate this." The system returns a set of signals. The seller picks the ones they want to act on. Adoption follows ownership. That is not a surprising finding. It is just how people work.

Myth 3: A seller can only close a million dollars per year

Von's internal research, conducted across 50 organizations in 2024, found that the average seller spends 1.2 hours a day in external meetings. The highest team average recorded was 2.4 hours. No organization in the study exceeded that figure.

Think about what sits behind those numbers. A junior seller in the Bay Area costs at least $200,000 a year in total compensation, running two or three customer meetings a day and spending the remainder rebuilding context between calls.

Sales carries a real cognitive load. Managing fifteen enterprise deals means holding in mind what was said in each last call, what to drive in the next one, and how to frame the business case for a specific buyer. Beyond the calls themselves, sellers are building decks, ROI models, and internal alignment. The time loss is not laziness; the task is genuinely demanding.

With AI that understands how a seller writes, what tone they use, and what the company's best-performing decks look like, the work that takes an evening compresses into minutes.

Hundreds of sellers are doing this today. Push external meeting time from 1.2 hours to four hours a day, and your annual quota can reasonably become a quarterly one.

The myth worth retiring is that a million-dollar annual number and 1.2 hours of daily customer time are an acceptable combination.

Myth 4: Build an agent for every task

Every organization is building agents. Meeting prep agents, deck-building agents, forecasting agents, and follow-up email agents. Agents designed to build other agents.

The trajectory is familiar. Think about what happened with dashboards. One person built one, then someone else built another, and over time the organization accumulated hundreds.

Dashboards broke when the business changed. The people who maintained them moved on. New hires built fresh ones without knowing the old ones existed. Today, most people work from three dashboards they trust and route everything else through RevOps.

Agents are following the same pattern. As the underlying data shifts and the company's structure evolves, point-purpose agents fall out of date. Nobody notices until something is obviously wrong, and by then the trust is gone.

The answer is a single agent that manages all the underlying agents on the user's behalf. Claude Code is not a collection of hundreds of separate tools. It is one entry point that handles everything, and when something breaks, there is one place to go.

Revenue teams need the same model: one agent for meeting prep, deck building, proposals, CRM updates, and everything in between.

Myth 5: More AI tools mean more intelligence

Every function in a revenue org is now being sold its own intelligence layer. CS leaders have one. Solutions engineers have another. AEs, Salesforce admins, and account managers each have options competing for their attention.

All of these tools draw from the same underlying sources. The CRM, the data warehouse, the call recorder, the engagement platform, and internal communications. Ask the same question across four different tools and you will get four different answers. The CS tool returns one number, the sales intelligence tool returns another, and neither knows what the other said.

Engineering avoids this because it has converged on one tool. The principle behind Salesforce's dominance was identical: every function inside one platform, working from shared data. Salesforce was never a point solution. Product marketing, admins, AEs, and CSMs all lived inside the same system and saw the same version of reality.

The answer for AI follows the same logic. One platform with one intelligence layer across the entire go-to-market stack, not a collection of point solutions pulling from the same data and returning contradictory answers.

A note on AI fatigue

Here's the analogy I'd ask you to sit with. AI is similar to electricity. Everything will be powered by it. Look around a living room. The TV. The refrigerator. The oven. The toaster. The Roomba. The fan. The lights. Even the couch was built using electricity. There isn't a couch in the world made without it.

AI will be in everything you touch. Do you walk into your living room and say there's too much electricity in here? Unless your bill is too high, hopefully sales pays well enough that you don't worry about it.

The fatigue you're feeling is something else. You've bought a lot of digital photo frames and Roombas. Nothing against Roombas. But you don't have enough TVs yet. You don't have enough applications creating the same kind of impact in your business that electricity creates in your living room.

Where this leaves you

Your raw data is already good enough. Sellers are spending 1.2 hours a day in front of customers, and your quotas should reflect what AI can unlock from the remaining six. Stop building agents for every task and invest in one. Stop accumulating point solutions and invest in shared intelligence.

The standard to aim for is simple: your sellers should be unwilling to go back.

How to make AI work as hard for revenue as it does for engineering