This week, in a thirty-six hour window, three posts crossed my feed saying the same thing from three different microphones. Aaron Levie, CEO of Box, tweeted on May 7: if everyone else does exactly what you do with the same technology, how do you stand out? Gergely Orosz, who writes the Pragmatic Engineer newsletter, tweeted the same day and racked up 1.6 million views in 24 hours: fast to build, but getting the world to know about it is increasingly hard. Eric Siu, founder of Single Grain, published a three-thousand-word essay on May 6 explaining, step by step, what to do when building costs zero.

What the three are saying, without coordinating, is the thesis I’m putting on the table this week with any founder or CTO of a B2B software company: the bottleneck moved up one floor. Building, the thing that used to slow companies down, is now the easy part. The next floor isn’t more LinkedIn ads. It’s something more specific: named distribution loops with named owners that learn each cycle. If you don’t have them, you don’t lose to a company with a better product. You lose to one with better loops.

The bottleneck moved up one floor

Three questions I ask when a founder tells me they’re using AI to accelerate their build:

What percentage of your team works on building the product versus delivering it to the customer? If the answer skews eighty-twenty toward building, you’re still fighting the 2018 fight. The asymmetry that made sense when building was expensive is operating debt today.

How many distribution channels did you kill this week? If the answer is zero, you don’t have a loop, you have a script. A loop without a kill rate doesn’t learn. It accumulates.

Who is the named owner of your demand-capture loop for the next ninety days? If the answer is “the marketing team” or “the CMO decides quarterly,” that isn’t ownership. It’s ambiguity with a budget.

This week’s thesis, simple: building got commoditized, the differentiation moved one floor up, and that floor has structure. It isn’t vibes. Eric Siu published the structure in his May 6 essay: seven loops and three layers. When I read it, I recognized the playbook AI Maestro has been running with clients for months, in different vocabulary.

Today’s piece is the downstream sister of Monday’s post on building got cheap, deciding didn’t. That one measured what’s scarce upstream of the build (what to build). This one measures what’s scarce downstream (who you deliver to, and how). Same argument: when a layer commoditizes, the differentiation moves up. Whoever takes it on now pays one; whoever takes it on in 2027 pays three.

Three microphones in thirty-six hours

Aaron Levie’s tweet on May 7 picked up 46.6K views: if everyone else does exactly what you do with this technology, how will you stand out? His thesis is single-sentence: when a capability becomes abundant, resources migrate to the layers that still produce differentiation. The companies that figure that out first capture the next two years; the ones that argue about it on Twitter for nine months pay for the same lesson with worse cards.

Gergely Orosz’s same-day tweet hit 1.6 million views, which on its own should make any B2B software founder pay attention. The line he kept hammering: the easier it is to build, the more the only differentiation is marketing and advertising. The pain isn’t shipping the product anymore. The pain is reaching the customer who cares.

Eric Siu published the operator’s manual on May 6. It’s not a tweet, it’s a three-thousand-word essay. The line that opens it marks the change in framing: agencies should stop selling labor; they should sell managed growth loops.

The chorus around the three posts is worth reading whole, not just the agreement:

• Zi Morris said it in one line: easy commoditizes, hard becomes the moat. The companies winning this cycle aren’t using AI to do the easy stuff. They’re redirecting the freed-up time toward the hard stuff AI can’t compress.
• Glitch Truth dropped the line with a number: distribution is the new moat; Cursor hit one hundred million in ARR not because the model is special but because they nailed dev mindshare before Copilot could react. That number alone is the proof for the thesis.
• Jatin Garg connected it to the Anthropic argument: AI compresses production, distribution becomes the moat for the same reason the agent runtime is the moat for AI labs. When the underlying primitive commoditizes, the layer above is the differentiation. Direct echo of the April 23 post.
• Ghali Bennis added the useful counter: sales, marketing, and support get commoditized by the same wave; AI SDRs, automated cold outbound, conversational support, all available to everyone in the same quarter. The real differentiation moves to what AI cannot compress: taste in what to build, and distribution you actually own.

That last line separates the people reading the wave from the people reading the headline. Distribution-you-buy is saturable. Distribution-you-build is defensible.

Siu’s three layers are AI Maestro in different vocabulary

Siu names three layers in his essay:

1. Managed loops — the owner of the outcome. The layer that decides what gets measured, what gets killed, what gets scaled.
2. Specialist agents — the workers inside the loop. Research, generation, QA, analytics, narrative.
3. Human judgment up top — named owner of taste, strategy, hard calls, what to kill, what to scale.

When I read his post on Wednesday, the first thing I thought was: that diagram is the same one we draw on the whiteboard in the first AI Maestro meeting with every client, just with different nouns. Where Siu says “loop,” we say named workflow with owner and kill rate. Where he says “agent,” we say agent with bounded blast radius and an evaluations pipeline. Where he says “human judgment up top,” we say decision rights and escalation SLA.

The overlap isn’t coincidence. We’re describing the same object from two verticals: he’s coming at it from a growth agency, we’re coming at it from B2B transformation. When two people who don’t know each other land on the same diagram from opposite sides, it usually means the diagram is pointing at something real.

The seven loops Siu names:

1. Raw materials — inputs that never go stale (customer pain, customer language, proof, offer, objections, channel data).
2. Creative testing — a hundred variants, kill ninety fast, learn from ten, compress the next batch.
3. Demand capture — find where attention lives, translate it into briefs and pages that convert.
4. Conversion — find where momentum dies between click and close.
5. Client narrative — turn performance data into decisions, not into recurring meetings.
6. Sales enablement — proof, objection handling, what’s proactive on every call.
7. Learning — capture winners and losers each cycle, feed the system, make it harder to replace.

The seventh is the moat. The first six can be imitated in six months. The seventh, the system that learns each cycle and stores what it learned in reusable format, takes years to imitate, and only if the organization redesigns around it.

For a B2B software company thinking about its next round or its next vertical, the operating question that comes out of the Siu/AI Maestro overlap is direct: how many of your seven loops are named with an owner, a kill rate, and a memory? If the answer is three, your moat is painted, not built. Technology Partner exists for companies that want all seven before the competitor reaches four.

Five distribution cycles since 1990. The bottleneck moves up.

I’ve been in computing for thirty-six years. Started at fifteen on a Commodore 64. I’ve watched this same move five times: a layer becomes cheap, the differentiation migrates one floor up, the industry spends three to five years complaining about the new bottleneck until the market puts a price on it.

Early 1990s — physical retail and trade shows. Reaching the customer was measured in booths, travel, and geographic sales force. The company that knew how to work a Las Vegas trade show paid one; the company showing up with a catalog and business cards paid three for the entire decade.

Late 1990s and early 2000s — telesales and email blast. A cheap channel appeared (mass mail, outbound calls to purchased lists), saturated in five years, regulated in seven with CAN-SPAM in 2003. Whoever understood it first built owned lists; whoever showed up late bought the list from a broker who’d already sold it to three competitors.

Mid-2000s through 2012 — SEO, SEM, and AdWords. The most scalable cheap channel the industry had ever seen. The companies that figured out keywords and CTR early built billion-dollar businesses; the companies that arrived in 2014 paid three times the same lead a competitor was buying for one in 2008.

2012 through 2020 — organic social and creator economy. Another cheap channel, another five-year cycle, another saturation, another regulator (iOS 14.5 ATT in 2021 broke Meta’s ROAS overnight).

2026 — owned loops and agent-readable surface. The cheap thing is building. The scarce thing is being readable by an agent in thirty seconds, having named loops with owners, retaining the customer through memory across sessions. Whoever has a demand-capture loop with a named owner and an in-house evals pipeline pays one in 2027. Whoever doesn’t pays three and maintains a Frankenstein of three rented channels.

Sequoia published the $1:$6 thesis (one dollar in software, six in services) in Q4 2025. Six months later, Levie, Orosz, and Siu confirmed it from three microphones in a single week. Cursor hit one hundred million in ARR while the conversation was happening. Earlier cycles spent five to seven years complaining about the new bottleneck before the market put a price on it. This one is already paying.

Five questions for Monday’s operations meeting

Before approving next quarter’s budget, take these five questions into the room. They’re not rhetorical. They’re the ones that separate the company paying one in 2027 from the one paying three.

1. What three distribution channels did you kill this week? If the answer is none, your kill rate is zero, you’re not learning. You’re running the same script you ran in January. A loop without kills is a script with makeup.

2. Who is the named owner of your demand-capture loop for the next ninety days? Not “marketing,” not “the CMO,” not “the committee.” First and last name, with authority to kill channels and authority to scale what works. If nobody passes both tests, the loop has no owner.

3. How much of your top-selling channel do you own and how much do you rent from Meta or Google? Rent is saturable and disappears when the algorithm changes. Ownership (list, community, customer memory, agent-readable surface) is defensible. The right ratio depends on the business; the question doesn’t.

4. Can your product be read in thirty seconds by an agent your customer is already using? If your customer asks Claude or ChatGPT “compare the three options for X” and your product doesn’t show up, you don’t have a marketing problem. You have a readable-surface problem. Wednesday’s post covered it.

5. How many PRs per week does your distribution team ship versus your product team? If product ships thirty and distribution ships two, you’re still investing in the layer that already commoditized. The right ratio isn’t one-to-one, but a twenty-to-one toward product means you’re ignoring the bottleneck that moved up.

At IQ Source we run AI Maestro before the next quarter goes into raising the model’s context window or hiring another outbound vendor. The audit maps the seven loops, names the owners for ninety days, sets the kill rates, and leaves an audit trail that survives the CMO change. For software companies whose product is part of the customer’s surface, Technology Partner extends AI Maestro into the product roadmap: distribution and product share the same vocabulary and the same owner.

Building costs zero. Distribution is the new investment. And like every investment, the first quarter with a named owner is worth more than the fifth quarter without one.