This is a different kind of post.
For the last thirteen pieces, we operated in research mode. We made claims, cited evidence, argued with data, and tried to be honest about what we didn't know. We never told you what we were doing about any of it. We were deliberate about that -- because the argument needed to stand on its own before we attached anything to it.
But there's context you don't have. Over the last 10 months, alongside the writing, we spent a significant amount of time talking to people: founders building in this space, GTM leaders running revenue teams, CROs who've lived through multiple cycles of tooling and consolidation, and friends and family who've sat across the table from buyers in ways that data alone can't capture. We weren't doing formal research interviews. We were having real conversations -- about what's actually broken, what the next three years look like, and whether the problem we kept writing about was real or theoretical.
We're still not pitching you anything. But we want to be transparent about something: those conversations didn't stay theoretical. Somewhere in the middle of them, the debate stopped feeling purely academic. The question shifted from "is this a real problem?" to "what would it actually take to solve it?"
This post is about what we believe -- the actual hypothesis. It's about why we think the current market of revenue tools, despite being worth tens of billions of dollars and despite being staffed by genuinely smart people, has a blind spot that nobody has solved. And it ends with a direct question we want your honest answer to.
First, a map of the existing landscape
Before you can understand what we think is missing, you have to understand what exists. Let me describe the current market honestly -- not as a competitor teardown, but as a structural analysis of what each category was built to solve, and what each was not.
There are roughly five archetypes in the current revenue intelligence market, and they have been consolidating fast. Gartner published its first-ever Magic Quadrant for Revenue Action Orchestration in December 2025, formally recognizing that previously separate categories -- sales engagement, conversation intelligence, and revenue forecasting -- are collapsing into each other. Two of the biggest players in the forecasting and engagement categories completed a merger in December 2025, combining approximately $450M in ARR. The CRM ecosystem's largest player is acquiring conversation intelligence capabilities and embedding them into its agent layer. The consolidation signal is clear: the market itself knows the point-solution era is ending.
Here is how I think about the archetypes and their honest limitations:
THE CURRENT MARKET -- WHAT EACH ARCHETYPE WAS BUILT FOR: ┌──────────────────────────────────────────────────────────────────┐ │ ARCHETYPE 1: THE CONVERSATION CAPTURE LAYER │ │ │ │ What it does: Records calls, transcribes, surfaces keywords, │ │ identifies coaching moments, tracks deal signals in meetings. │ │ │ │ What it was built for: Making call review more efficient. │ │ │ │ What it cannot do: Retain memory across rep transitions. │ │ Cross-reference signals against product usage, email, support. │ │ Act on what it captures without a human reviewing first. │ │ Build a compounding model of a relationship over years. │ │ │ │ The honest limitation: Intelligence surfaced to a dashboard │ │ that waits for a human to look at it. The loop never closes │ │ autonomously. │ └──────────────────────────────────────────────────────────────────┘ ┌──────────────────────────────────────────────────────────────────┐ │ ARCHETYPE 2: THE FORECASTING ENGINE │ │ │ │ What it does: Pipeline analytics, deal health scoring, │ │ roll-up forecasting, rep accountability dashboards. │ │ │ │ What it was built for: Giving revenue leaders more confidence │ │ in their number before the board call. │ │ │ │ What it cannot do: Close the loop between "this deal is at │ │ risk" and "this action was taken to address it." Understand │ │ why deals were lost in human relationship terms. Connect │ │ pipeline health to marketing signals or product signals. │ │ │ │ The honest limitation: Brilliant at prediction. Quiet on │ │ prescription. Mute on execution. Requires manual override │ │ at every step. │ └──────────────────────────────────────────────────────────────────┘ ┌──────────────────────────────────────────────────────────────────┐ │ ARCHETYPE 3: THE CRM ECOSYSTEM │ │ │ │ What it does: Everything, in theory. CRM, CPQ, billing, │ │ conversation insights, AI agents, revenue analytics. │ │ The ambition is a single system of record for all of revenue. │ │ │ │ What it was built for: Enterprise lock-in through ecosystem. │ │ │ │ What it cannot do: Deliver conversation intelligence at │ │ the sophistication of a dedicated platform. Run agents at │ │ the depth required for complex B2B relationships. Operate │ │ at a price point available to non-enterprise orgs. │ │ │ │ The honest limitation: A $500--650/user/month stack that │ │ produces generalist AI against specialist problems. The │ │ conversation intelligence is less mature than competitors. │ │ The agents are primarily oriented toward B2C workflows. │ └──────────────────────────────────────────────────────────────────┘ ┌──────────────────────────────────────────────────────────────────┐ │ ARCHETYPE 4: THE SALES ENGAGEMENT PLATFORM │ │ │ │ What it does: Sequences, cadences, multi-channel outreach, │ │ SDR workflow automation, pipeline execution at top-of-funnel. │ │ │ │ What it was built for: Helping SDRs send more touchpoints │ │ faster, with higher consistency. │ │ │ │ What it cannot do: Build post-sales intelligence. Understand │ │ a customer relationship's history. Connect outreach context │ │ to what's happening in the account post-signature. │ │ │ │ The honest limitation: Top-of-funnel muscle with no memory. │ │ The signals generated by engagement never compound into │ │ organizational intelligence. Each sequence starts from zero. │ └──────────────────────────────────────────────────────────────────┘ ┌──────────────────────────────────────────────────────────────────┐ │ ARCHETYPE 5: THE AI-NATIVE NEWCOMERS │ │ │ │ What it does: Autonomous agents for specific revenue │ │ workflows -- prep briefs, follow-up drafts, MEDDIC scoring, │ │ meeting summaries, next-step recommendations. │ │ │ │ What it was built for: Replacing the most annoying manual │ │ tasks in a rep's day. │ │ │ │ What it cannot do: Build shared organizational memory. │ │ Operate across Sales, Marketing, and Product simultaneously. │ │ Close the loop on actions that span the whole revenue cycle. │ │ │ │ The honest limitation: Point solutions with AI inside. │ │ Replaces one tool in a stack of fifteen with a smarter tool. │ │ Doesn't restructure the stack. Doesn't build the substrate. │ └──────────────────────────────────────────────────────────────────┘
The structural gap none of them fill
Read those five descriptions carefully and ask yourself a question: which one of them builds a continuously compounding, organizational-level model of every customer relationship -- one that spans Sales, Marketing, and Product -- and then acts autonomously on what it learns?
The answer is none of them.
This is not a failure of ambition. Each of these platforms was built to solve a real problem for a specific buyer. The conversation platform was built for sales managers who needed more efficient call review. The forecasting engine was built for CROs who needed better visibility. The CRM ecosystem was built for enterprises who needed one throat to choke. The sales engagement platform was built for SDR teams that needed scale. The AI-native newcomers were built for individual reps who were drowning in admin.
Every one of them was designed with a single-function buyer in mind. None of them were designed around the thesis that Sales, Marketing, and Product are all revenue-responsible functions operating on the same customer -- and that fragmentation between them is not a cultural problem to be managed but an architectural problem to be solved.
The result, as Forrester's 2025 State of RevOps survey found, is that 58% of B2B companies cite process misalignment as the primary barrier to growth -- not technology, not talent, not market. Internal fragmentation. Despite ten years of "alignment programs," the problem has not moved. That is what happens when you apply process solutions to architecture problems.
The hypothesis
Here is what we believe.
The revenue technology category has optimized furiously for the wrong constraint. The constraint everyone has been optimizing is: how do we make humans faster and more informed? Better dashboards. Better summaries. Better coaching. Better forecasts. Better briefs.
The constraint nobody has adequately addressed is: how do we build a system that perceives everything happening across every customer surface, builds compounding organizational memory from those signals, and acts autonomously on the decisions that don't require human judgment -- so that humans are reserved exclusively for the moments where judgment is the actual product?
Those are not the same problem. They produce completely different architectures.
The first approach builds better tools for humans to use. The second approach builds an intelligent substrate that the organization operates on.
A better dashboard still requires a human to review it, decide what it means, and then act. The insight-to-action gap -- the window between a signal firing and something being done about it -- remains intact. In most revenue organizations today, that gap is measured in days to weeks.
An intelligence substrate closes the loop. Signal fires. System processes. Autonomous action executes -- on the defined class of decisions that don't require a human. Human is alerted only when genuine judgment is required. The gap collapses. The model learns from every outcome.
THE INSIGHT-TO-ACTION GAP -- WHERE REVENUE DIES: CURRENT STATE: Signal fires ──► Dashboard updates ──► Human reviews (days later) ──► Human decides ──► Human schedules action ──► Window often closed Revenue lost in the gap: immeasurable but consistent ────────────────────────────────────────────────────────────────── HYPOTHESIS: Signal fires ──► System processes ──► Autonomous action (defined class) ──► Human alerted only for judgment-required situations Revenue recovered from the gap: structurally, not heroically
The second architecture is not science fiction. It's what the market is slowly converging toward -- just piecemeal, one category at a time, without anyone designing the whole.
We think someone needs to design the whole. That's the question we keep coming back to.
What "memory" actually means in this context
One word in the hypothesis deserves more precision: memory.
We don't mean storage. We don't mean searchable transcripts or a better CRM record. Every platform in the market already logs. Logging is not memory. Logging is the raw material of memory that has never been synthesized.
By memory, we mean a continuously updated model of a customer relationship -- one that captures not just what was said, but what was meant; not just the deal stage, but the political dynamics inside the account; not just the product usage, not just the email thread, not just the support ticket -- but all of them, synthesized into a living representation of the relationship that belongs to the organization, not to the rep.
That model should survive rep turnover. It should inform a new rep in hours, not months. It should update when the buyer changes roles, when the champion leaves, when the budget cycle shifts, when a competitor is mentioned in a call by a different team on a different surface. It should connect what Marketing knows about this account to what Sales experienced and what Product observed. And it should act on what it learns -- not wait to be asked.
Network World's summary of Gartner's top strategic prediction puts it sharply: "In this new ecosystem, verifiable operational data becomes a currency, fueling a data feed economy where digital trust frameworks and verifiability are prerequisites for participation."
The organizations that build this compounding intelligence layer now are building a moat. Not because the software is hard to replicate -- software is always replicable. But because the organizational learning that accumulates from thousands of interactions, acted on and refined over years, cannot be reproduced by buying a license. It compounds with time and cannot be purchased backward.
Now, the 36--60 month question
We want to ask you something directly.
Gartner's top strategic prediction for 2026 and beyond -- presented to thousands of CIOs and IT leaders at their annual symposium -- is this:
"By 2028, 90% of B2B buying will be AI agent intermediated, pushing over $15 trillion of B2B spend through AI agent exchanges."
-- Gartner Strategic Prediction 2026 (source)
That's not a 10-year horizon. That's 24 months from now.
The implication is structural: the interface of B2B commerce is changing. The buyer's AI evaluates your positioning before your rep is contacted. The shortlist is formed in a query, not a nurture sequence. The vendor who wins is the vendor whose signal is clearest, most verifiable, and most structurally consistent -- not the one whose rep sent the best sequence.
In this world, organizational memory isn't a nice-to-have. It's the substrate. Your buyer's AI is building a model of you. Your revenue system should be building a compounding model of your buyer -- one that gets smarter with every interaction, acts autonomously on everything it learns, and reserves human attention for the moments that actually require a human.
| Gartner Prediction | Timeline |
|---|---|
| 90% of B2B buying AI agent intermediated | By 2028 |
| $15 trillion B2B spend through AI agent exchanges | By 2028 |
| 80% of customer-facing processes via multi-agent AI -- dominant orgs | By 2028 |
| 22% of monetary transactions programmable for AI agent economic agency | By 2030 |
These are not fringe predictions. They come from the same firm whose forecasts drive enterprise software budgets worldwide. As Digital Commerce 360 noted: "Gartner analysts issued 10 strategic predictions for 2026 and beyond. They warn that AI is accelerating faster than many organizations can manage."
The market is already signaling the shift. The Clari--Salesloft merger combined two dominant platforms in December 2025 -- a forecasting engine marrying a sales engagement layer. Salesforce is acquiring conversation intelligence companies to close the gap between insight and execution. Every major player is racing to own the action layer, not just the insight layer. The convergence validates the thesis: insight without autonomous action is a dead end.
So here is the direct question
We believe B2B buying is going to be fundamentally different in 36 to 60 months. Not incrementally different. Structurally different. The mechanism of evaluation, shortlisting, and decision-making is being rebuilt from the ground up, and the revenue organizations that are still running on insight-first, human-executed architectures will face a compounding disadvantage that gets harder to close every quarter.
We are exploring what it would take to act on this belief.
Not a CRM. Not a coaching tool. Not a better dashboard or a smarter copilot. The thing worth exploring is an intelligence substrate -- a system that perceives signals across every customer surface, builds organizational memory that compounds over time, acts autonomously on the decisions that don't require human judgment, and reserves your best people for the moments where their judgment is the actual product.
That's the hypothesis. That's what we're interrogating.
But we want to know what you think -- because if we're wrong about the premise, we need to know now, not after anyone commits to building it.
Three questions we're genuinely wrestling with:
- Do you believe B2B buying will be structurally different in 36--60 months? Not "AI will be more common" -- but structurally different in how evaluation, shortlisting, and purchase decisions happen?
- If you accept that buyer behavior is changing at this rate, does the gap we described -- between insight and autonomous action, across Sales, Marketing, and Product on a unified memory layer -- feel like a real constraint to you? Or does your current stack feel like it's converging toward solving it?
- Is this worth solving for? Is the problem acute enough -- and is the window real enough -- to justify exploring it seriously now rather than watching the incumbents close it?
We don't have all the answers. We do have a thesis, a conviction, and enough early signal to keep asking the question. The whole reason we've been writing this series publicly is that the thinking is better when it's contested.
If you've read this far, you're probably the right person to contest it.
If this series has been useful to you -- or if you think we're fundamentally wrong about any of it -- we'd genuinely like to hear from you. Not as a sales motion. As a conversation between practitioners who care about getting this right.
You can find us here. The debate continues.