For years, we bought software the way kids buy LEGO.
You bought the fire station. You bought the hospital. Each one came in its own box, already glued together, with instructions for exactly how you were meant to play with it. That was SaaS. The finished set was the product. The box was what you paid for.
Underneath all of it sat a different thing entirely. The big bin of plain bricks. The data, the rules, the permissions, the plumbing. The raw material you can build anything out of. That was infrastructure.
Two different stores. One sold you finished toys. The other sold you the bricks. For a long time those were separate businesses, and everyone knew which one they were standing in.
That wall is coming down.
The clearest sign came last week, when Databricks shipped CustomerLake, an agentic CDP sitting right on top of its own bricks. I started this conversation on LinkedIn, and it clearly hit a nerve, because the comments turned into the exact argument the whole industry is about to have.
Who Owns the Bin? Who Owns the Set? (Wrong Question)
Now that the wall is down, everyone is lining up to fight about who wins.
Does the infrastructure company eat the application company. Do apps become thin skins stretched over data platforms. Who owns the customer. Who owns the workflow. Who owns the data.
It is a real debate, and smart people are having it. Scott Brinker, who has tracked this industry longer than almost anyone, has been circling the same idea for months: that the value is quietly moving to context, to the system that actually understands how your business works. I am not the first person to say this out loud, and I will not be the last.
But here is the catch the whole layer-war keeps missing.
You can win the argument about who owns the bin and still lose. Because the brick that decides whether a deal closes was never sold in either store.
It was never a brick at all.
Meet Maya: The Person Your Whole Pipeline Actually Runs On
Maya runs demand gen at an $80M cybersecurity company. Six-figure deals, nine-person buying committees, two-quarter procurement cycles.
Her stack is excellent: Salesforce, Marketo, 6sense, Gong, Clay. Dashboards for everything.
But here's what everyone on her team knows: the dashboards are not why the pipeline moves. Maya is.
She looks at an account lighting up in 6sense and says ignore it: that's their research team, not a real signal. She tells an AE to throw out the deck because she saw the buyer just came from a company burned by a competitor. She knows which case study closes which objection because she's heard the same objection seventeen times.
None of that lives in a dashboard. It lives in Maya.
Then Maya goes on parental leave. Four months. Same tools, same team, same budget, same playbook docs in Notion.
The quarter comes in soft. Conversion slips. Deals stall in legal. A forecast that looked clean in week two unravels by week ten.
Her team had every tool she had. They were missing the one thing that made the tools work.
She comes back. Within a quarter, it recovers. Nothing in the system got fixed, because nothing in the system was ever broken. The context was missing for a while.
I've lost count of how many versions of Maya I've watched. The names change. The pattern does not.
Her 'Instinct' Was Just a ‘Data-Join’ No System Was Built to Do
Here is the part the infrastructure debate keeps missing.
Maya's judgment did not live in her head because it was unwritable human magic. It lived there because no system she owned could hold the whole funnel at once. So she became the integration layer, in human form.
When she "just knew," she was cross-referencing the 6sense spike against the Gong call against the LinkedIn job change against seventeen objections in her memory, fast enough that it felt like instinct. It was not instinct. It was a join across data the company already had.
The job-change signal existed in a tool. The recurring objection existed, buried in call recordings. The account heating up existed in the intent platform. Every brick was already in the city, sitting in fifteen different boxes that never talked to each other. Maya was the only thing connecting them.
That is the uncomfortable truth underneath "context lives in people." A lot of what we romanticize as irreducible human judgment is really just integration that no machine was ever positioned to do.
Is some of it genuinely tacit? Yes: taste, timing, the read on a relationship, the decision to walk away from a deal that looks great on paper. That part stays human, and it is exactly where you want your best people spending their judgment.
But that is the thin slice. The thick part, the part that actually degraded the quarter Maya was out, was pattern recognition across signals the company was already paying to collect. That is not a human problem. It is a systems problem.
Your Agents See One Box. Maya Saw the Whole Board.
This brings me back to the thesis that prompted this newsletter.
Because we are trying to solve a human problem with more agents. You hand an agent one tool's slice of the world and ask it to act. It writes a faster cold email. It summarises a call. Useful, not transformative.
A single-tool agent is just a faster version of the fragmented stack that made Maya necessary. It can see its one box. It cannot see the board.
The warehouse-native wave is a real step. Pull everything into one governed place and you finally give agents a substrate to stand on. But a data lake is not the same as operational context. The warehouse holds raw material. It does not know that the champion went dark, the objection pattern matches three deals you lost last year, and the right move is a specific play run by a specific person this week.
Owning the bin gets you the bricks. It does not get you the map.
Five Steps to Pull the Join Out of Someone's Head, Starting This Week
Here is the part most pieces like this skip. You do not need to buy anything to start fixing this. You need to make the invisible join visible, and none of what follows requires a new platform.
1. Start with the leave test.
Go stage by stage through your funnel and ask: if the person who "just knows" went dark for a quarter, which decisions would quietly get worse? Write the names down. Most teams find two or three Mayas they did not realise they were depending on.
2. Write down what those people actually join.
For each one, get specific about what they cross-reference when they "just know." It is almost never mystical. It is a short, nameable list: leadership changes, real intent versus noise, stalled deals, quiet champions. Name the actual signals.
3. Find where each signal already lives, and where it dies.
Map each signal to the tool it already sits in: enrichment, intent platform, call recordings, CRM. You will almost always discover the data exists; it just sits in separate boxes nobody joins. The intuition was the join.
4. Before you touch any tooling, agree on the words.
Lock shared definitions: what counts as qualified, at risk, stalled. Most funnels run on terms that mean different things to marketing, sales, and finance, which is why dashboards disagree and handoffs leak. It costs nothing and everything else stands on it.
5. Put the named signals on one screen, even crudely.
Before any AI or new platform, build one shared view where those signals sit together for your top accounts, and one weekly ritual where the team looks at it and decides together. A spreadsheet that pulls from your tools counts. The day a junior rep can make Maya's call because the screen shows what Maya saw, you have captured the context.
That is the manual version. It is slow and held together with exports, and it will not scale. But it does two things immediately: it de-risks you from losing your Mayas, and it tells you whether the leverage is big enough to make the join permanent. That is the next decision.
Once You've Found the Join: Build It, Buy It, or Automate the Trap
The market has named the job of making the join permanent: GTM engineering. Postings grew 205% last year, with top salaries north of $250K, because the work is exactly this: turning scattered go-to-market signal into a system that sees the whole funnel and acts on it.
The old answer to a soft funnel was headcount. That math broke. B2B teams now spend roughly two dollars in sales and marketing to earn a single dollar of new ARR. You cannot hire your way out of that. You have to build leverage, which leaves three options.
Build the integration layer yourself. | Activate a platform that already is one. | Buy Point Agents and run a fragmented stack |
Total control, and the right call if your motion is genuinely unusual or your data is real IP. Just know the first working version takes months of engineering time, and then you own the maintenance forever. | You get the join, identity resolution, and the intelligence layer pre-built. Point your team at pipeline instead of plumbing. This is the bet we made at RevSure: a full-funnel data graph that unifies signals into one model of how your revenue moves, with agents that run on that whole graph. If you want the honest build-versus-activate math, including when building yourself is right, we wrote that up here. | You buy five single-purpose agents from five vendors and call the pile a strategy: one for outbound, one for enrichment, one for scoring, none sharing a brain. That is the fragmented stack all over again, except now it is autonomous and much harder to see into. |
Whichever path you pick, be honest about one thing: we are not capturing the irreducible part. Nobody is bottling taste or a read on a room, and you should not want them to. That stays human. What gets captured is the thick, unglamorous middle, the integration that was never anyone's highest use of judgment anyway.
The Teams That Win Won't Pick the Right Layer - They'll Close the Context Gap
Here is what the layer-war fundamentally misses.
The LEGO wall is coming down because the wall stopped mattering. Once a machine can assemble bricks on demand, owning the bin or the finished set becomes a footnote. Every layer collapses into a single question: can your system see what your best operator sees.
That is not an infrastructure question. It is not even a software question. It is a context question, and the only one that survives the collapse.
You do not need RevSure to start answering it. Find your Mayas, name the signals they are joining by hand, and drag that join onto a screen your whole team can use. Do that and you have already taken most of the ground. You have turned a person into a system. Whether you then build the rest, buy it, or keep running it by hand is a real decision with real tradeoffs. But it is one you get to make from strength.
The teams that win the next few years will not be the ones who picked the right layer. They will be the ones who closed the gap between what their people know and what their systems can act on.
That gap is the whole game now.
Everything else is just bricks.
Know a revenue leader who is secretly the integration layer for their entire team? Forward this to them. They will recognise themselves.
P.S. You can run the leave test this week without us, and you should. But if you find a join too big to hold together with spreadsheets, that is exactly where a platform earns its keep. That is why we built RevSure. Reply here, or find me at [email protected].