Next week, a room of 100+ of revenue leaders are going to walk out with an AI agent workflow they built themselves, already doing real work. I mean the kind of workflow that runs while you're in meetings. The account brief is waiting before the call. The pipeline report lands in your inbox Monday at 7 without anyone pulling it. The follow-up after an event goes out the same day. Most teams I meet already pay for the tools to do this. Copilot, Gemini, ChatGPT or Claude is sitting on every laptop. What's missing is time set aside to build something real against the one constraint that's slowing the number down. So that's the session. I'm running the Part 2 build session after Matt Heinz's workshop, 90 minutes, for sales, marketing and RevOps leaders. You pick your biggest GTM constraint, and you build the workflow that attacks it, in the tool your company already approved. You'll leave with it running, and with a kit of 33 more plays to keep going when you get back to the office. All at the at @6sense Breakthrough Agentic GTM event. Let me know if you are going to be there below Session time and room are in the first comment.

If your fiscal year starts in January, your 2027 number is being built right now. Probably in a spreadsheet. Probably on last year's assumptions. THE CRO CALENDAR: annual plan. Late October to mid November. CRO, CFO, RevOps. Here's how I'd rebuild planning season with AI. BEFORE THE FIRST PLANNING MEETING → Prompt: rebuild capacity from actuals, not targets. Real ramp, real win rate by segment, real cycle length, from the last six months of clean data. No benchmarks. → Prompt: list every assumption in this year's plan next to what actually happened. → Skill: model the plan three ways. Current productivity, and two cases where AI gives reps back selling time. Show which case the number depends on. IN THE ROOM Three decisions only the CRO can make: 1. Which segment do we bet on? 2. Which assumption are we least sure of? 3. What would make us change the plan in March? AFTER → Routine, monthly from January: compare actuals to every plan assumption and flag the first one that breaks. The goal isn't a plan that survives the board meeting. It's knowing in March which assumption broke, while there's still time to act on it. I've trained 5,000+ executives on AI. This is where it pays first: the plan everything else hangs on. Full build in the first comment: schedule, connections, and the exact instructions to paste. Which assumption in your 2027 plan has been tested the least?

How I'd audit every AI automation running across a revenue team with one prompt. I just ran this on myself. 150 automations across GitHub, Claude, ChatGPT, and my own laptop. Here is the exact prompt: "Give me a map of every automation I'm running across Claude Code, GitHub, and locally on this computer. I have too many automations and many of them overlap. Build it in a Google Sheet titled Automation Master Mapping. Organize it by platform and by category so I can see which ones overlap. Do not change or delete anything. Then give me your recommendations based on everything you see. Patterns, what to consolidate, and the opportunities I'm missing." The prompt does the mapping. The insight comes from four questions: 1. Where does it live? → Cloud agents, scheduled runs, repos, local scripts → If you can't list them, you can't govern them 2. What job does it do? → Group by category, not by tool → Overlap only shows up when you sort by job 3. Which model was it built for? → Anything written around an old model's limits is dead weight → The new model may do it in one step 4. What's missing? → The map shows what's running → The gaps show what should be The most important line in the prompt is the one that says do not change or delete anything. That line is the difference between an audit and an incident. NOTE: The goal is not a cleaner stack. It is knowing which automations actually move your number. Read first, retire second, build third. In that ORDER. You can't consolidate what you can't see. You can't see what you never mapped. Which question would your team stall on first?

How I’d audit a CRO’s week for automation in 10 days. Based on the audit I am running on myself right now. Stop asking “where does my time go?” Start asking “which of my 50 hours this week actually needed me?” Here is the 10-day playbook: DAY 1: INSTALL THE WATCHER → A screenshot every 15 seconds → 10 to 15 questions a day, answered in the moment → Decide what stays off camera before you start DAYS 2-6: CAPTURE A REAL WEEK → No cleaning up because it is recorded → Answer the why, not just the what → Leave the messy parts in DAY 7: COUNT THE REPEATS → Cluster the screenshots into recurring tasks → Hours at stake = times per week x minutes each → Mark tasks that exist only because two systems do not talk DAY 8: SCORE THE JUDGMENT → 1: same steps every time → 3: rules with exceptions → 5: only you can make the call DAY 9: SORT INTO THREE PILES → Low score, high hours: automate → Middle scores: delegate with a written playbook → Score of 5: keep it, and protect the time DAY 10: START THREE → One automation, one handoff, one thing you stop doing → Each gets an owner and a date NOTE: The goal is not a cleaner time log. The goal is a business where you stop being the DEFAULT for every task. I have trained 4,500+ executives on AI. The scoring sheet and report template are in the first comment. Your calendar shows your meetings. Your screen shows your job. Next week I will share how you can take the next step and start automating.

Four years ago I was on a plane and asked three questions that took ten years off something. I told this story again on Thursday, to a room of executives at a distributor, and I noticed I still get slightly embarrassed telling it. The context: I had spent about a decade learning a particular piece of go-to-market craft the long way. Repetition, mistakes, deals lost while I worked out what I had got wrong. That decade was expensive and I was proud of it. Then I had early access to an early model, a bad one by today's standards, and I asked it three questions on a flight. The answer came back about 95% of the way to where ten years had taken me. My first reaction landed somewhere closer to grief. Then irritation, then about an hour of wondering what the decade had been for. The useful reaction took longer. What that session actually proved was that the knowledge part of the craft had become cheap, and the knowledge part was never the hard part. Knowing what to ask a buyer is available to anyone now. Sitting in silence after you ask it is not. Reading a room, carrying a committee, telling a sponsor something they will not enjoy hearing: none of that got cheaper that day. I went all in that week, and the four years since have mostly confirmed the split. The teams that struggle with AI are usually protecting the part that got cheap. The teams that move are the ones who handed it over and spent the returned hours on the part that did not. I still would not trade the decade. It is what lets me tell when the model is confidently wrong, which is the one skill this whole field runs on and nobody puts on a certificate.

"I get AI envy every time I talk to him." A sales leader told me that on Wednesday about a friend of his who runs a similar business. Same size. Same private equity backing. Same kind of problems. His friend talks about the board decks his systems assemble and the insights that arrive before he asks for them. Then this leader goes back to his own week, which involves a CRM his team treats as paperwork and a model his CTO picked because it came bundled. His phrase for his own situation was "the dark ages." I hear this more than almost anything else now, and it is worth naming what is actually happening, because it is not a technology gap. Both of these men have access to the same models. Neither has a technical team building anything custom. The friend runs an ordinary stack. The difference is about eighteen months of continuous small decisions. Which workflow to attack first. Who owns it. What to measure. What to stop doing. None of those decisions required a budget approval or a vendor. Eighteen months is also roughly how long the gap takes to become uncomfortable to describe at a dinner. It compounds quietly and then shows up all at once, in a peer conversation, as envy. The useful thing about envy is that it is specific. He could name exactly what his friend had that he wanted, which means he already knows what to build first. Most of the leaders I work with arrive after a conversation like that one. Something a peer said made the gap concrete. If you are the one being envied right now, your advantage is eighteen months of decisions, and it is still compounding while the other person works out where to start.

"We have been relying too much on our AEs to generate top-of-funnel opportunities. It's really all AEs at this point." A revenue leader at a marketplace business said that to me on Monday, describing a 2027 plan to double enterprise pipeline. Here is the arithmetic sitting underneath it. Each of his enterprise AEs carries just under 200 accounts. Success for one of those reps looks like closing one deal a month. So the plan asks a rep who owns 200 accounts, and who has to close twelve deals a year, to also personally create twice the pipeline. With the same headcount. Nobody in that room thought the plan was crazy, me included. They all knew exactly what it needed, which is why the conversation went where it went. Three places top-of-funnel can come from, and most enterprise teams only staff one: The AE. Highest conversion, lowest volume, and every hour spent here is an hour not spent closing. This is the expensive option and the one most teams default to. The SDR team. Volume, and a scoring problem. Without a signal that says which of the 200 accounts is in market this month, an SDR works the list alphabetically and burns the territory. Marketing. The only one that compounds, and the slowest to show a number. Also the one that gets cut first when the quarter is tight. The move his team landed on was to skip the debate about which lever to pull and answer a narrower question first: which of those 200 accounts are in market right now. I can get that answer from public signals in an afternoon, and having it makes the other two levers worth pulling. Doubling pipeline with the same headcount is a prioritization problem before it is a capacity problem. If you had to double top-of-funnel next year with the team you have today, where would the first new opportunity come from?

"Those prompts, even though we taught them how to use it, are pretty much untouched." A change management lead at a 10+ Billion company said that to me on Thursday, in front of his own AI council, about a training series she personally finished delivering. That takes some nerve. It is also the most accurate sentence anyone said on the call. Her team did everything the playbook says. Built the series. Taught the prompts. Shipped the library. Then went back and looked at usage, which most teams skip, and found the library sitting there untouched. What struck me was her diagnosis, not the finding. She is redesigning the next series around a gap she named precisely: there was no time, no space and no permission to fail. Her words: "there's time and space and energy to work with one another, practice the prompts, and fail without any kind of repercussion. And it's specifically saying, leadership is saying, this is important." This is the the step 1 FOUNDATION for transferrable results. Four ingredients in that sentence, and a prompt library has none of them. Time means it is on the calendar during work hours, not homework. Space means a room with peers, not a seat at your own desk. Failure means the first attempt is expected to be bad and nobody logs it. Leadership means somebody with a title says this matters before anyone is asked to try. A prompt library is a reference book. Nobody learned to sell from a reference book. The teams I see get real adoption all did the same unglamorous thing: they put a recurring practice block on the calendar and a leader in the room, and they let the first three attempts be rough. If you have shipped a prompt library this year, go look at the usage data before you ship another one. The number will tell you which problem you actually have.

A client handed me a number this week that I am going to use in every AI business case from now on. $3,750. That is what one hour of one of his sellers is worth. He got there with arithmetic a finance team cannot argue with. Take the team's quota. Divide by the total selling hours available to that team in the year. The result is the revenue potential of a single selling hour. He called it selling potential per hour, and he built it himself because he needed something his CFO would accept without a debate about whether AI is real. Here is why that one number changes the conversation. Every AI business case I see argues about the tool. Cost per seat, model quality, whether the pilot proves anything. Those arguments have no end, because nobody in the room can price the thing being saved. Put a dollar figure on an hour and the argument inverts. Now the question is how many hours of assembly work sit inside the week, and what it would be worth to move even a fraction of them back to selling. How to run it for your own team this week: Take one team's annual quota. Divide by the selling hours they actually have, after travel, internal meetings and admin. Write the number down. Then open your last time study, or run a five-question one, and find the hours going to research, prep, follow-up and CRM updates. Multiply. That product is the size of the prize, in the only currency your finance team reads. A caveat I would say out loud in the room: an hour returned is NOT revenue until somebody sells with it. The number sizes the opportunity. Your managers decide whether it converts. This means reallocating focus based on newly available time. Those are the steps steps before the revenue number starts to move through implementing AI tools

A Sales Leader at a $1B+ company asked me the most useful question I have heard this year. We were 35 minutes into a call with his AI council, his change management lead and two business unit heads. He had been quiet the whole time. Then he said, "Please excuse the direct question, but it's the thing I've been thinking the whole time." "How is this different or better than us having Copilot teach us what it can do?" Most vendors dodge that one. I answered it straight, because the honest answer helps him either way. Copilot can teach a seller what Copilot does. Deciding which of that seller's 17 weekly jobs gets handed over first along with change management is a different problem, and making their manager care whether it happened is a third one. He summed it up better than I did. "Copilot can absolutely do that, but the structure mechanism in it is the struggle." Three thousand sellers each working it out alone gives you three thousand private experiments, running at the speed of whoever had a slow week. What a structure supplies that a license does not: 1. A ranked list of jobs to hand over, by role, not by feature. 2. A time and place to practice, on the calendar, with the manager in the room. 3. A measured before and after, so the first result arrives as a number. 4. A named owner per workflow, so nobody waits on the council. His closing line was the sharpest part of the call. "So the failure point is still kind of, they still need to do this, and the change management still needs to be important to the manager." He is right. That is where I would spend the first 30 days.

OpenAI shipped dots on September 29. An always-on ChatGPT agent that keeps working between conversations, in Slack and Teams too. Most people will brief it like a search box. The brief that works is one sentence: "Get me [outcome] by [date]. Use [source]. Done means [test]. Ask before [action]." Five real jobs: 1. Forecast prep → Commit list I can defend, by Friday at noon → Source: call transcripts and email, not the CRM stage field → Done means: every deal over $100K shows the last economic buyer contact. Ask before changing a close date. 2. Pipeline scrub → Keep, push, or kill on every stalled deal, by Monday 8am → Source: the last 30 days of email and notes → Done means: one line of evidence per verdict. Ask before messaging a rep. 3. Renewal risk → Accounts renewing in 120 days, ranked → Source: the signed contract, not the original proposal → Done means: each flag cites a usage or ticket trend. Ask before contacting the customer. 4. Account brief → One page per top 20 account, before QBR week → Source: two quarters of calls plus order history → Done means: a rep runs the meeting without opening another tab. 5. Inbound reply → First reply drafted within 15 minutes of a demo request → Source: our approved messaging doc → Done means: one proof point that fits their industry. Ask before sending anything. NOTE: The dot does not get smarter between job 1 and job 5. The brief does. One outcome, one SOURCE, one test, one line it cannot cross. I have trained 5,000+ executives on this. Prompts describe the work. Briefs define when it is finished. Which of these five could your team not define done for?

"The decision to go with Copilot is a lazy one." A CRO at a $50m+ company said that to me last week. He added straight away that he did not intend it as an insult to anyone. By his read it was a bandwidth decision. He had asked his leadership for different model licences. His CTO, in his words, "is buried and doesn't have the bandwidth to consider my requests," and came back with the answer every CTO gives when there is no time to run an evaluation: we already pay for Microsoft, you already have Copilot. That answer is defensible on its own terms. It also sidesteps the question he actually asked. I want to be careful here, because I work with plenty of companies that standardiszd on Copilot and get excellent results. The model is rarely the thing that decides whether a rollout works. What matters is that nobody made a decision. A default got applied under time pressure and now has the status of strategy. Two months later, his team is running what he calls a skunkworks, trying to build something useful on a tool nobody chose, while he works out how to reopen a conversation that was never really opened. If you are the leader holding the short end of a default like this, here is what I told him to do. Stop arguing about which model. You will lose that argument, because your CTO has no time to referee it and the incumbent is already paid for. Argue about one workflow instead. Pick the single highest-value thing your team does badly today. Build it on what you have. Measure it. Then build the same thing on the tool you wanted, and put the two outputs side by side. Now the request has changed shape. You are not asking for a new platform. You are showing a measured gap on one named workflow, with a cost next to it. A CTO with no bandwidth cannot evaluate a platform. He can read one comparison. What default is currently running as strategy in your go-to-market stack?

We built an AI prospecting agent for a Fortune 500 sales team. The first feedback session was blunt. Good. Three things broke. None of them were the model. 1. THE DATA RAN OUT. The agent pulled contacts through each rep's own data-provider account, and those accounts were nearly spent. The fix was a shared team login, not a new model. 2. IT WAS BUILT FOR One Segment of Accounts. The logic was tuned for mid-size companies. The reps were pointing it at $1B+ accounts, where you multi-thread by title and map relationships before anyone gets a message. It kept writing to VPs when the buyers were directors of manufacturing and quality. 3. THE MESSAGES WERE 80% THERE. A good rough draft, the reps said, but nothing you'd send a top account unedited. The fix: we're feeding it the emails that already worked for their best reps. The target is 95%. One rep said it surfaced account signals well and wrote tailored, punchy messages. Another said it didn't fit their top accounts at all. Both were right. THE PRE-ROLLOUT CHECKLIST FOR ANY PROSPECTING AGENT: 1. Whose data credits does it use, and how many are left? 2. Which account tier was the logic built for? 3. Which titles actually buy, by tier? 4. On top accounts, does it map relationships before the first message? 5. Has it read the five best emails your top reps ever sent? The model was the easy part. The last 20% lives in your reps' sent folders. There's definitely a different motion between that new pipeline and the existing pipeline. With a lot of nuances to take it to a fully automated solution for the last 10% Or the last mile

A sales leader at a founder-led company doing about $8M told me what finally convinced their leadership team that AI was real. It wasn't a sales use case. It was a back-office job handled by seven to ten part-time contractors. Unreliable, slow, and always a little late. One AI workflow replaced the whole process in 48 hours. That convinced the leadership team. They stopped asking whether AI works and started asking what else it could do. It's the pattern I keep seeing. The first win that changes minds is almost never the one on the roadmap. It's the annoying task with unreliable handoffs and a deadline nobody enjoys. THE FIRST-WIN FILTER (pick the task that passes all four): 1. Someone already complains about it out loud. 2. It repeats every week, so the saving compounds. 3. "Good" is easy to recognize on sight. 4. A skeptic can watch it work within two days. Skip the strategic use case in round one. It needs six approvals and a committee. Win first with the task nobody defends. Then earn the right to go bigger. What's the task in your company that everyone complains about and nobody has automated?

A sales leader at a founder-led company that is doubling every year told me they had built AI tools for their own proposals, emails and contracts. They work. They can't get them to the team. That is what "adoption is ad-hoc and uncoordinated" looks like up close. Not a lack of tools. Tools that only work for the person who built them. The reason is almost always the same: → The prompts live in one person's head or their personal skill/assistant directory → The context lives in one person's account → "Good" is a taste nobody wrote down So the team sees the output, calls it magic, and can't repeat it. The cleanest fix I've seen came out of one of our Shark Tanks this summer. A team at a travel-tech company built a "mini-me" for each teammate: an assistant trained on a short write-up of how that person works. Now anyone can draft an email sequence or a social post in a colleague's style. THE PORTABILITY TEST (4 questions, 5 minutes): 1. Could a new rep run this workflow on day one without asking you anything? 2. Where does the context live? If the answer is "my chat history," it doesn't travel. 3. Who fixes it when it breaks? A name, not a shrug. 4. Is "good" written down as an example someone else can compare against? Fail two and you have a personal tool, not a team workflow. The fix is boring. Move the prompt into a shared file. Move the context into a shared folder. Attach one before-and-after example. Name an owner. Which of your best AI workflows would still run if you took a week off?

Every AI project I build, for a client or for myself, starts with the same file. Not a prompt. A context document. It's the fastest fix I know for "we rolled out Copilot, and reps use it like Google." The model doesn't know your business, so it answers like a search engine. It can't sell what it can't see. Custom instructions run out of room fast. A context document doesn't, which is why it now opens every demo I give. WHAT GOES IN IT (one document, plain text): 1. Who you are: what you sell, to whom, at what size. 2. Your buyers: titles, what keeps them up at night, the words they actually use. 3. Your offers: what each one does, and what it costs the buyer not to have it. 4. Your proof: results you're allowed to quote. 5. Your voice: three examples of writing that sounds like you. 6. What's off limits. HOW TO USE IT: → Upload it as a source to every project → Update it when an offer or a proof point changes, not when you remember → Split it when you serve very different buyers. One team I worked with had a single document written about the parent company. Every division needed its own. If your AI read only one document about your business, what would it get wrong? This single step alone will take you from an advanced Google search engine to creating top 1% outputs.

This might be a pretty big deal

Last week in New York, I sat in a 175° sauna, then dropped straight into a 40° cold plunge. I stayed in the water for two and a half to three minutes. The sauna was sensory overload. Heat on every part of you and nowhere for your attention to go. The plunge flipped all of it in one breath, and for those few minutes the only thing I could focus on was the next breath. When I climbed out, I felt more refreshed than I have in a long time. A conference week is noise from the first coffee to the last dinner, and this was the one stretch where my head went completely quiet. Captivate Talent hosted it during the GTM conference. Huge thanks to @Chris Gannon and the whole team for building something this unique into a week full of panels and badge scans. Also, it was great plunging with @Cliff Simon and @kendre morales and also chatting with @RJ Belles and @Will Polese while having our own personal shaman.