The Budget Shift Spiegel Predicted Is Real. Most of That Money Is About to Be Wasted.

Evan Spiegel went on the TBPN podcast and said: companies used to pour resources into engineering, and now "people are going to be much more focused on marketing, on distribution." He isn't theorizing; on a separate podcast this spring he noted that more than two-thirds of new code at Snap is now AI-written, and that "Claude is transforming software development, full stop." A brand agency telling you a tech CEO just declared marketing the new bottleneck is, obviously, suspect, which is exactly why this post exists: to agree with Spiegel on the evidence, and then to correct the conclusion everyone's drawing from him, because the popular version of this take is about to waste an enormous amount of money.

The shift is real, and we invoice it

The evidence stack for the production collapse is short and brutal. Andrej Karpathy, a co-founder of OpenAI, coined "vibe coding" in early 2025 for describing what you want in plain English and letting the model build it; within months, Y Combinator reported that a quarter of its Winter 2025 batch had codebases that were roughly 95% AI-generated. Snap's two-thirds figure says the same thing at public-company scale. And our own pricing page says it in dollars: development of a finished one-page site at Space runs $5,000 in about half a week, $15,000 for 25 pages in about a week, prices that exist because agents carry construction and QA while a senior engineer owns judgment. We didn't predict software's commoditization. We repriced it and published the number.

So yes: when building stops being the constraint, the money moves to whatever is. Spiegel says that's marketing and distribution. Half right.

The half everyone's missing: vibe marketing is on the same curve

Here's the logical error in "software commoditized, therefore marketing is the moat": the same models that write the code write the campaigns. Infinite ads, infinite landing pages, infinite content calendars, generated as easily as the app they promote. Ninety-four percent of marketers already plan to use AI for content creation, engagement rates are falling across every major platform as the feeds flood, and the fatigue is measurable enough that researchers now link the decline in social media time directly to low-quality AI content. If "anyone can build it" kills software as an advantage, then "anyone can generate it" kills marketing-as-production by the identical argument, on the identical timeline.

The bottleneck didn't move sideways from engineering to marketing. It moved up, in both functions at once: from production to judgment. In engineering, review and architecture became the job while typing became free. In marketing, the same inversion is arriving on schedule: deciding what to say, to whom, with what evidence, became the job while saying it became free. Companies that read Spiegel as "hire a content team and buy reach" are about to fund both sides of a slop war. The ones that read him correctly will fund the layer above it.

What the reallocated budget should actually buy

Four things, none of which is volume. Positioning: one sharp, provable claim, dug out of customer evidence, because in a market where everyone can ship, what you're known for is the product decision that matters most. Distinctiveness: the sameness tax compounds precisely as production floods, and looking like your category now costs more than it ever has while departing from it costs less. The signature stack: named humans, receipts, and accountability, because as anonymous perfection approaches free, provenance is where the value concentrates. And the answer layer: distribution in 2026 isn't only feeds and app stores; it's whether AI engines recommend you when a buyer asks, a channel where the unit of competition is one citation, not a thousand impressions, and where almost nobody is measuring their position yet. That last one is literally checkable in 24 hours for free, which makes it the cheapest first move of the entire reallocation.

There's an organizational echo too. Anthropic's head of growth recently suggested the PM-to-engineer ratio may need to flip as engineering leverage explodes; TIME just documented tech companies renaming marketing roles into engineer-flavored titles to make the shift internally palatable. Different rooms, same signal: judgment functions are absorbing the headcount and status that production functions are shedding. We wrote the job description for where that lands.

The honest limits

Three of them, in fairness to the engineers. Software isn't zero moat: compute scale, proprietary data, network effects, and genuinely hard systems still defend companies, and Snap's own two-thirds figure describes new code, not the infrastructure underneath it. Vibe-coded prototypes aren't production systems: the practitioners' honest reports converge on the last 10% (auth, rate limits, maintenance) being where the free lunch ends, which is why our own model pairs agents with senior engineers rather than replacing them. And the messenger caveat, stated plainly: an agency profits from this thesis, so don't take our word for the shift. Take Spiegel's budget behavior, YC's batch data, and the fact that we cut our own development prices to commodity levels in public, which is not the move of a firm pretending code is still scarce.