The Year Software Ate Its Own Tail

---

We’re watching the reckoning happen in real time, and it’s uglier than the usual startup funeral.

The pattern’s unmistakable if you’ve been paying attention: companies that built on hype and weak fundamentals are getting crushed, while the infrastructure problems nobody wanted to solve are eating everyone’s lunch. The question isn’t whether this is a correction. It’s whether anyone actually learned anything.

When Scale Becomes a Liability

Broadcom acquired VMware in 2023 and immediately started raising prices. The response? Thousands of migrations to competitors. Not thousands of customers considering other options. Thousands actively running for the exits. That’s not customer dissatisfaction—that’s a vote of no-confidence executed at the infrastructure level, where migrations cost real money and real pain. You don’t rip out your virtualization layer because you’re grumpy about pricing. You do it because the alternative became less scary than staying put.

This is what happens when a company mistakes market position for actual defensibility. VMware had lock-in, sure. But lock-in only holds if customers believe you’ll treat them reasonably. The moment Broadcom signaled it wouldn’t, those locked-in customers became latent demand for alternatives. That’s a dangerous position to be in—the market was just waiting for permission to leave.

Photo by Pixabay / Pexels

The Infrastructure Collapse Nobody’s Talking About

Meanwhile, actual security is melting down and we’re treating it like background noise.

Iran-linked hackers disrupted operations at US critical infrastructure sites. Russia’s military hacked thousands of consumer routers. Nvidia GPU clusters are now completely vulnerable to Rowhammer attacks that give attackers full control. And OpenClaw—whatever that is—apparently gave everyone another reason to panic about security. These aren’t theoretical vulnerabilities in beta software. These are active, weaponized compromises of systems millions of people depend on.

Here’s what gets me: we’ve built this entire AI stack on commodity hardware, cloud infrastructure, and consumer-grade networking. The same routers getting compromised are sitting in office buildings where companies are training their large language models. The same GPUs that Rowhammer can fully control are what everyone’s betting their compute strategy on. We’re not just ignoring security. We’re building the future directly on top of known, exploited vulnerabilities.

The worst part? There’s no company making money by fixing this. Security is a cost center. AI agents are a growth story. Guess which one VCs fund.

The Talent Vote of No Confidence

Doug Field just left Ford. For those not tracking automotive talent moves: Field was Apple’s chief vehicle platform officer, then Tesla’s VP of engineering, then Ford’s EVs and tech chief starting in 2021. If you’re building cars that run software, you want Doug Field working on them.

Ford didn’t keep him. And the auto industry should be sweating.

What this actually signals is that the Big Three auto executives don’t know how to retain the people who actually understand how to build vehicles that compete with Tesla. They can offer stock options and titles, but apparently not the operational freedom or strategic clarity these leaders need. Field didn’t leave because Ford’s pay wasn’t competitive. He left because the job wasn’t winnable.

That’s a structural problem that money doesn’t fix.

Photo by UMA media / Pexels

Why Startups Are Getting Acquired By Caterpillar

Monarch Tractor built an electric tractor with autonomous driving capability. It sounded great in a pitch deck. The reality: dealers hated it, farmers complained the tech didn’t work, and a co-founder said the same thing publicly. Now Caterpillar owns it—not because Monarch proved the market, but because Caterpillar can absorb a failed bet and strip it for parts.

This is how the consolidation era works. You don’t acquire startups because they’re winning. You acquire them to bury their technology roadmap, neutralize their founders before they build a real competitor, or occasionally salvage specific technical talent. Caterpillar buying Monarch isn’t validation that autonomous electric tractors work. It’s evidence that they don’t—yet—and Caterpillar wanted to ensure nobody else cracks it first.

The startup that sounded revolutionary six months ago is now a line item on a quarterly earnings call.

The AI Agent Play (For Real This Time)

OpenAI updated its Agents SDK. On the surface: a toolkit update. In context: the bet that “agentic AI”—systems that make decisions and take actions autonomously—is where the actual revenue will eventually come from.

Hightouch is proof the model might actually work. They built an AI agent platform for marketers and grew from $30M to $100M ARR in 20 months, adding $70M in that window. That’s not a hockey stick. That’s a rocket curve. And it’s specifically because they built on top of AI agents, not despite it.

My read: OpenAI’s SDK update is them building the picks-and-shovels layer that Hightouch was forced to build themselves. If OpenAI succeeds in making agent-building accessible and standardized, companies like Hightouch have a much faster path to scale. OpenAI gets a percentage of everything built on top of their platform. Everybody wins except the companies trying to build agents from scratch without that leverage.

This is actually how technology consolidation should work—faster platform adoption, better developer experience, lower friction. Too bad the rest of the year’s news is companies doing the opposite.

The Live Nation Question

A federal jury found Live Nation acted as an illegal monopoly. The company then settled with the DOJ. But here’s where it gets weird: settlements aren’t the same as breakups. Live Nation keeps existing as Live Nation. The settlement probably involves behavioral restrictions, maybe some divestitures, definitely some fines. But the core question—should this company actually be broken up—remains unanswered.

I think the headline’s sarcasm is justified. Breaking up Live Nation would require actual structural dismantling of a company that’s spent two decades becoming the only option for touring musicians and venue operators. The political will for that doesn’t exist. So we’ll get a settlement, a few years of compliance theater, and then Live Nation will figure out the new game within the constraints they’ve been handed.

It’s the corporate equivalent of being fined for speeding when you own the highway.

Photo by Denys Gromov / Pexels

The Actual Story Here

This year we’re seeing the difference between scale and strength. Broadcom has scale but lost strength. VMware’s customers voted with their feet. Monarch Tractor had a cool idea but no execution. Ford can’t keep talent. The infrastructure is actively compromised and nobody’s funding the fix. And the only growth story that’s actually working is the one where founders built the entire platform stack themselves before anyone else proved the concept.

My prediction: we’re about to see another round of consolidation, this time focused on companies that built real execution moats rather than market-share moats. The next 18 months will be brutal for anything that can’t prove it actually works and that customers actually prefer it. Hype is finally getting expensive.

The companies that’ll win are the ones that look boring next to what failed this year.

What I’m Watching

• Broadcom customer migration velocity through Q2 2025. If the migration rate accelerates beyond “thousands,” we’re looking at a genuine VMware replacement cycle, not just edge cases. Watch for which competitors gain the most share—that tells you where the market actually thinks the better product is.

• The next critical infrastructure attack and the response time to patch. If we see a major compromise of Nvidia-based infrastructure using Rowhammer, and it takes more than 30 days to deploy fixes, we’ve officially got a national security problem dressed up as a vendor issue.

• Whether OpenAI’s Agents SDK produces more Hightouch-scale successes by end of 2025. If they don’t, the agent platform bet wasn’t actually solved—it was just outsourced to the companies already winning.

• Ford’s next five EV/software leadership hires. If they keep losing talent at the Doug Field level, the company’s EV strategy is dead regardless of their product roadmap.