Before you point AI at your pipeline, decide where the truth lives

TL;DR: I recently ran a project entirely inside an AI assistant and watched it slowly fall apart, not because the AI was bad, but because I'd asked it to be both the worker and the record. Moving the record into Salesforce fixed it: the AI does the work, the CRM holds the truth. If you're exploring AI for your sales operation, that division of labour is the decision to get right first.

I rebuilt my website recently, and I started the project the way plenty of people now start projects: inside an AI assistant. I gave it all the context, all the files, all the background. At first it was brilliant: it remembered everything. Then, a few rounds of amendments in, things got messy. The assistant was reading several versions of the same information and they'd started to conflict. I spent longer arguing with it about what was current than I spent on the actual work.

The fix wasn't a better prompt. The fix was giving it a database. I moved the project's record into Salesforce, and now my workflow has a clear division of labour: the AI is the engine for doing the work, and Salesforce is my database and my source of truth — always up to date, always maintained.

The conventional view

Most conversations about AI adoption are conversations about tools. Which assistant, which model, which licence. The CRM, in this framing, is yesterday's technology, the legacy system the shiny new thing will eventually replace.

The challenge

An AI assistant's memory is personal and conversational. It's shaped by what you told it, in the order you told it, and it lives with you. That's wonderful for one person on one task, and it quietly breaks the moment a second person is involved. If you've spent the morning updating an AI on a customer, your colleague's AI knows none of it. Two people working with the same assistant are not necessarily working from the same facts.

A CRM is the opposite. Connecting your AI to Salesforce is almost like putting a shared database in the middle that you can both talk to. You work independently, but you're reading and writing the same record — so when you log a conversation at ten, your colleague sees the latest position at two. The AI stops being a private notebook and becomes a front door to something the whole team can trust.

The strategic truth

The division of labour that works is simple to state: AI does the work, the CRM holds the truth. The assistant drafts, summarises, chases, and surfaces what needs attention. The system of record stays structured, shared, and current underneath it.

Two things follow from that. First, consistency becomes something you can design. In my own business, the AI works through predefined skills, which means every project, every task, every update lands in the same structure: my team works the way I work, not five different ways. The conversational interface is friendly; the record behind it is disciplined.

Second, automation needs a stopping point, and the right one is usually just before the decision. On a recent project we designed the renewal process so the system assembles the renewal record automatically when a deal closes, and a person then clicks approve, because a human glance at the dates catches what automation can't know. Automate the draft, keep the decision. Your team stays in charge of the judgement calls; the admin burden quietly disappears.

One honest caveat, because it matters: none of this is magic. Getting AI to work well against a CRM depends on the record being worth reading: fields that mean something, a structure that matches how you actually sell, data the team genuinely maintains. That foundation is real work, and anyone who tells you otherwise is selling something.

Why this matters to your business

If your customer truth currently lives across inboxes, spreadsheets and individual memories, AI doesn't resolve that, it answers confidently from whichever fragment it happens to see. The ambition is right; the sequencing is everything. A business that sorts its source of truth first gets an assistant that makes the whole team faster. A business that doesn't gets plausible-sounding answers built on partial data, and adoption dies the first time someone catches the AI confidently wrong.

What to do about it

Ask one question of your own operation: if a customer rang right now, where would the person answering look for the current truth? If the answer is one place, you're ready to do interesting things with AI on top of it. If the answer is "it depends who you ask", that's the place to start. Not as a grand programme, but as a deliberate tidy-up of the record AI will read. Then pick one workflow (meeting notes into the account record, a daily list of what needs attention) and let the assistant earn its place.

The engine is only half the machine

Everyone's choosing engines right now. The businesses that pull ahead will be the ones that also built something worth connecting the engine to.

If you're weighing up what AI could genuinely do on top of your Salesforce and what needs to be true of your data first — that's a conversation we have a lot at the moment. Book an Agentforce conversation and we'll map it against how your team actually works.