How to Turn ChatGPT Into Your Sales Operating System with Alex Miora

Most salespeople are still using AI one prompt at a time.

• Ask. 
• Answer. 
• Copy. 
• Paste. 
• Repeat.

Useful? Sure… But if that’s all you’re doing with ChatGPT, you’re leaving most of the value on the table. The bigger opportunity is connecting AI to the places where your sales context already lives: email, calendar, CRM, Drive, Slack, meeting transcripts, and delegating entire workflows instead of isolated tasks. That was the experiment in a recent Predictable Revenue Podcast episode with Collin Stewart and Alex Miora, GTM at OpenAI. Instead of another conversation about what AI might do for sales someday, they built around three practical workflows: a current deals dashboard, a Follow-Up Friday system, and living account plans. The result points to a bigger shift in how sales teams should think about AI: The future of AI in sales isn’t AI doing the selling, it’s AI doing everything around the selling.

Your AI sales stack is probably a mess.

• ChatGPT. 
• Claude. 
• Codex. 
• n8n. 
• CRMs. 
• Spreadsheets. 
• Docs. 

Automations built six months ago that nobody remembers how to maintain. Sound familiar? Each tool solves something. Eventually, though, the seller becomes the integration layer, moving data between systems, updating fields, rebuilding context before calls, checking three places to figure out what happened with one account. That was Collin’s starting point. His sales and AI workflows worked, but they were fragmented. So the question wasn’t, “What new tool should I buy?” It was: can one AI layer orchestrate the tools and data I already have? Alex’s answer was essentially yes. He describes ChatGPT Work as a kind of digital brain: connect it to the systems where your data lives, give it context about how you work, and let it handle the repetitive operational layer around selling. AI probably isn’t killing the CRM, but it may kill the way salespeople use it. At scale, companies still need systems of record. But the rep may no longer need to spend half their life clicking around inside them. The agent can increasingly handle the updates, retrieval, and maintenance while the seller focuses on decisions and customers.

Stop giving AI tasks. Start giving it workflows.

Most ChatGPT usage still looks like this:

1. Summarize this call.
2. Write this email.
3. Research this company.

Those are useful tasks, but they’re isolated. A workflow looks more like this: Understand my accounts, monitor what changes, update the relevant records, tell me what deserves attention, and prepare the next action. That’s a fundamentally different relationship with AI. A useful way to think about the shift is: Connect → Contextualize → Delegate → Review.

1. Connect: Give AI access to the relevant sources.
2. Contextualize: Teach it your methodology, preferences, definitions, templates, and sales process.
3. Delegate: Give it recurring outcomes, not just one-off prompts.
4. Review: Keep humans involved where judgment and customer-facing communication matter.

That framework is a synthesis of the workflow Collin and Alex build through the episode. And the three examples make it concrete.

Workflow #1: Build a sales dashboard that updates itself

Collin’s first goal was simple: stop hunting for the current state of every deal. He wanted one view that could answer:

• What deals am I working on?
• What changed?
• What happens next?
• Which opportunities have no next action?
• How qualified are they?
• Which ones actually deserve attention?

The important part is where the dashboard gets its context.

Step 1: Connect the sources where customer context lives

For most sellers, that means email and calendar first. Then Drive, CRM data if you have one, call transcripts, and other connected systems. Instead of manually keeping another database current, the AI can pull context from the places where the work already happened. That removes one of the oldest problems in sales operations: the CRM only knows what the rep remembered to type into it.

Step 2: Define the decisions you want the dashboard to support

• “Build me a CRM” is a bad instruction.
• “Show me active opportunities, next actions, missing next steps, deal stage, qualification, timeline, and risk” is much better.

The difference is subtle but important. You’re not asking AI to recreate software, you’re telling it what decisions you need to make.

Step 3: Optimize for action, not aesthetics

Collin makes a useful point while building the dashboard: he likes Kanban boards visually, but he doesn’t love them for running a sales process because they can hide the details he needs to make decisions. He prefers a sortable list where he can quickly see next-action dates, stage, and MEDDPICC scoring. That’s the lesson: Don’t build the dashboard that looks coolest. Build the one that tells you what to do next.

Workflow #2: Automate the follow-up you keep forgetting

Then comes Follow-Up Friday. The lazy version of AI follow-up is obvious: generate emails for everyone who hasn’t heard from you recently. Collin’s intended workflow is more useful: have the agent review the pipeline, identify the five highest-value follow-ups, explain why those deals deserve attention, draft personalized outreach, and leave the final communication for human approval.

• That changes the question from: Who hasn’t received an email lately?
• To: Who actually needs attention?

Alex takes the same idea further with his own forecast automations. His system can pull context from email and calendar, update deal information on a recurring schedule, flag meaningful changes, and eventually surface things he may have missed. Don’t automate follow-up, instead, automate the decision about what deserves a follow-up. The writing is the easy part. Prioritization is where intelligence becomes valuable.

Workflow #3: Build account plans that don’t die in Google Drive

Account plans have a predictable lifecycle: 

1. You create one. 
2. It looks great. 
3. Everybody feels very strategic.

Then reality happens: 

• New meetings. 
• New stakeholders. 
• Company announcements. 
• Leadership changes. 
• Product launches. 
• Strategic shifts.

A month later, the account plan is basically historical fiction.

Alex’s approach? Making the account plan an always-current asset instead.

1. Start with a reusable template. 
2. Give the AI access to internal context from email, Slack, calendar, CRM, Drive, or wherever the account history lives. 
3. Schedule it to check for relevant external news and update the plan when something actually changes.

That turns an account plan from a static document into a living intelligence layer. And this isn’t just useful for enterprise AEs.

• Founders can use it for strategic customers.
• Customer success teams can use it for expansion accounts.
• Agencies can use it for client plans.
• Consultants can use it for engagements.
• SDRs can use it for target accounts.

Your account plan should help you decide what to do next.

The best AI sales automation still has a human in the loop.

Alex draws a pretty clear line between work that can become highly autonomous and work that should stay close to a human.

1. CRM updates? Automate them.
2. Forecast maintenance? Automate a lot of it.
3. Research and monitoring? Let the agent run.
4. Customer-facing communication is different.

Alex still wants to be involved in follow-ups because the strategic decisions matter: who gets included, how something is phrased, what you ask for, what you offer, and what the account actually needs next.

Automate the grunt work, but never automate giving a damn.

That philosophy shows up in how OpenAI’s own SDR function works too. Despite having access to frontier AI, the company isn’t simply replacing SDRs with bots. Alex describes how one of their SDRs uses deep, agent-generated account context and research to make human outreach more relevant. The AI helps build the context corpus, then the human decides how to use it.

The real upgrade: give every seller more mental RAM

Salespeople are limited by time and context. Thirty opportunities, hundreds of emails, calls, stakeholders, competitors, internal conversations, next steps, industry news, etc. Nobody can keep track of it all.

Collin calls this the limit of human “RAM.” 

Alex’s answer is to use AI as a persistent context and ideation partner. Something that can keep track of what’s happening, pull together information across systems, and help identify what the seller is forgetting.

AI’s biggest sales advantage might be memory.

Alex shares an example where his system surfaced a major strategic blocker he had missed, something buried across external news and emails. He also uses a simple end-of-week question: what did I drop the ball on? And the system usually finds something. That’s a much higher bar than “draft this email.” You’re asking AI to challenge your blind spots.

• What am I missing in this opportunity?
• Who should I be engaging?
• What changed at this account?
• What context from our last conversation should matter now?

AI becomes more valuable when it stops being a vending machine for content and starts acting like an always-on second brain.

Start here: build your own AI sales operating system.

You don’t need to rebuild your entire sales stack tomorrow, just start with one recurring workflow that annoys you.

1. Connect your context: Start with email, calendar, Drive, CRM, and wherever customer conversations live.
2. Pick one painful recurring workflow: Pipeline review, follow-up, account planning, forecasting, meeting preparation. Aka grunt work.
3. Describe the outcome in plain English: You don’t need to turn prompt writing into a second career. Alex’s approach is increasingly simple: explain what you want, run it, and iterate.
4. Run it manually first: See what the agent does before making it autonomous.
5. Correct it: Teach it what “good” looks like in your workflow.
6. Automate it once you trust the output: Recurring work is where the leverage compounds.
7. Then ask the dangerous question: What am I still doing manually that you could be doing for me?

That’s when AI stops being another tool in the stack and starts becoming the operating layer across it.

The goal is to build an AI layer around a great salesperson.

One that remembers the details, watches the pipeline, updates the records, researches accounts, catches dropped balls, prepares follow-ups, and keeps the context current. So the human can spend more time doing the thing sales has always depended on: Talking to customers. That’s the irony of AI in sales. The more effectively you automate the work around the relationship, the more time you create for the relationship itself. NO TIME TO READ?