Hey — It’s Nico.

Welcome to another Failory edition. This issue takes 5 minutes to read.

If you only have one, here are the 5 most important things:

• Kyte, a startup that delivered rental cars to your door, has shut down — learn why below.

• The State of Vertical AI (Q2)

• Google’s AI Mode expands globally, adds new agentic features.

• FieldAI raises $405M to build universal robot brains.

• An MIT study says that 95% of AI pilots in companies lose money, but its not AI’s fault — learn more below.

If your startup sells to big companies, check today’s sponsor: Delve helps startups finish their SOC 2 process in 15 hours (not 15 weeks). And they’re offering a $1k off promo using code FAILORY1k!

Delve just raised $32M to kill compliance busywork.^AD

Delve’s on a mission to give millions of founders, engineers, and operators their time back by eliminating the endless busywork of compliance.

They recently announced a $32M Series A at a $300M valuation led by Insight Partners, with participation from top CISOs and Fortune 500 execs. They’re already trusted by 500+ high-growth companies (Lovable, Bland, 11x) and have doubled revenue in the last quarter.

If you’ve ever lost an enterprise deal because SOC 2 or HIPAA wasn’t in place, you know how painful compliance can be. It’s hours of screenshots, spreadsheets, and chasing evidence across Slack threads.

Delve’s AI agents handle all of that in the background, adapting to your systems and automating the hardest parts from start to finish.

Their customers have unlocked billions in pipeline and saved tens of thousands of hours — and now, with fresh funding, Delve’s scaling their platform even faster.

Skip the months of compliance busywork — book a demo, use FAILORY1k, and get $1k off.

This Week In Startups

🔗 Resources

We must build AI for people; not to be a person.

Why your AI product needs a different development lifecycle.

Glue teams vs back-office teams.

The State of Vertical AI (Q2)

📰 News

Google’s AI Mode expands globally, adds new agentic features.

SoftBank makes $2B investment in Intel.

‘Crazy conspiracist’ and ‘unhinged comedian’: Grok’s AI persona prompts exposed.

GPT-5 is supposed to be nicer now.

💸 Fundraising

FieldAI raises $405M to build universal robot brains.

IVIX’s software raises $60m to help governments fight financial fraud.

Nuclear energy startup Aalo raises $100M in funding.

AI agent startup TinyFish raises $47 million.

Fail(St)ory

The best competitor to Hertz?

This week, Kyte, a startup that delivered rental cars to your door and claimed to be “the best competitor to Hertz”, shut down. 

After a year of cuts and city exits, missed loan payments let its main lender repossess and liquidate the thing that made the whole model work: the cars. It’s a clean reminder that in asset-heavy businesses, your real competition isn’t just incumbents—it’s interest rates, lender terms, and time.

What Was Kyte:

Kyte set out to make renting a car feel like ordering food delivery. You booked on your phone, a contract driver dropped the car at your doorstep and picked it up when you were done. Unlike peer-to-peer marketplaces, Kyte owned and controlled the fleet. 

The founders believed professional fleets would beat peer supply on safety, consistency, and cost. Long term, they talked about a fleet that could transition to autonomy. With that pitch, the company expanded to 14 U.S. markets, stacked equity and credit lines, and marketed itself as a serious challenger to legacy rental brands.

Growth masked a harder truth: running a distributed, owned fleet is capital-hungry and operationally unforgiving. By 2024, the company was struggling to produce free cash flow in several cities and attempted a concentrated push toward profitability in San Francisco and New York. The turnaround didn’t land.

Then the unraveling. First came a quiet asset sale—Kyte handed its customer list to Turo. Next, the lender, triggered by missed payments, started taking back cars and selling them off. Once the fleet began disappearing, the product effectively vanished. The board looked for a lifeline, but with the inventory gone, the ending was already written.

The Numbers:

• 📅 Founded in 2019.

• 🏙 Operated in 14 U.S. markets at peak.

• 💰 Raised $300M+ across equity and credit facilities.

• 🔁 2024 restructuring: exited most cities to focus on SF & NYC.

• 🔚 Sold customer list and other assets to Turo; then shut down.

Reasons for Failure: 

• Asset-Heavy Model in a High-Rate World: Owning cars soaks up cash: buying or leasing, maintenance, insurance, parking, the works. With borrowing costs higher, interest bills ate into every rental. When revenue wobbled, loan terms tightened and the lender had the upper hand. 

• Delivery Layer Made Margins Thin: Kyte ran two ops at once: keeping cars utilized and running a courier service to drop-off/pick-up vehicles. That only works if cars stay busy and handoffs are tight. In many cities, trips per car stayed too low, and courier time + idle time crushed margin. Complexity scaled faster than utilization.

• Financing as a Single Point of Failure: Concentrating debt with one main lender is efficient—until it isn’t. Miss a few payments and that partner can pull levers you can’t. Once cars get taken and sold, revenue, reliability, and customer trust fall in a hurry.

• Principle Over Optionality: Kyte rejected peer-to-peer supply on principle. That removed a lighter, flexible way to adjust inventory for slow weeks, seasonality, or new city launches. A hybrid approach might have bought time while owned-fleet economics matured. Ironically, the customer list ended up with the very marketplace they dismissed.

Why It Matters: 

• Owning assets is a debt bet. Spread lender risk, leave covenant slack, and model slow, expensive refinancing.

• Scale only where units win. Prove seasonal, courier-included margins in one city before opening the next.

• Optionality beats dogma. Blend owned fleet with marketplace/partners to flex supply and keep utilization high.

Trend

95% of AI Pilots Fail

MIT just threw cold water on the AI party: its new study says ~95% of corporate pilots aren’t moving the P&L. Wall Street noticed — AI-heavy stocks slid and the Nasdaq dipped the next day. The vibe turned from “look, a cool demo” to “show me the earnings.” 

Why It Matters:

• The hype tax just got priced in. Big AI stocks wobbled as investors questioned near-term ROI (Palantir −9.4%, Nvidia −3.5%; Nasdaq −1.4%).

• It’s not the model—it’s the org. MIT’s authors talk about a “learning gap”: pilots don’t connect to real workflows, so they don’t pay.

• Strategy is shifting to plumbing, not stunts. Coverage shows better outcomes from buying proven tools and integrating—versus shiny bespoke builds that never ship.

What just happened

MIT Media Lab’s NANDA team looked at interviews with leaders, an employee survey, and hundreds of public deployments. Their punchline: only about 5% of gen-AI pilots show rapid, measurable impact. The rest stall in “pilot purgatory.” 

On the very same day, Sam Altman was telling reporters that “when bubbles happen, smart people get overexcited about a kernel of truth,” and when asked if investors were overexcited about AI, he replied: “my opinion is yes.”

Put those two headlines together and it’s not hard to see why the market panicked. That combo spooked traders, and AI stocks took a hit—Nvidia slipped, Palantir tanked almost 10%, and Arm dropped too. It wasn’t a meltdown, more like a reality check: the AI gold rush might not pay out as fast as people hoped.

If you’re thinking “but everyone’s using AI,” you’re not wrong. McKinsey’s latest survey shows 78% of orgs used AI in at least one function in 2024. Usage is up; enterprise-level impact is the bottleneck. 

Why the wheels come off

Pilots don’t fail because GPT forgot grammar. They fail because companies bolt a clever model onto a creaky process and call it transformation. MIT’s authors call this the learning gap—tools and teams aren’t learning from each other or from the actual workflow, so the system never improves where money is made. In short: great autocomplete, zero process change. 

Then there’s the buy vs. build trap. In-house experiments sound heroic; in practice, they drown in data plumbing, security reviews, and legacy integrations. Reports summarizing the MIT work highlight higher success rates when companies buy a battle-tested product and wire it in, instead of rolling their own app. 

Also, expectations got weird. We celebrated chatbots for writing emails faster, then expected them to reshape EBIT in a quarter. McKinsey’s read is more sober: business-unit wins are showing up, but more than 80% of companies still don’t see a tangible enterprise-level EBIT impact from gen-AI yet. That’s not doom; it’s a timing problem. 

Practical Optimism, not Theater

The MIT study isn’t anti-AI; it’s anti-theater.

Proofs of concept don’t move earnings—shipped workflows do. And when programs mature, returns are there: Deloitte reports that 74% of companies say their most advanced gen-AI initiative meets or beats ROI targets, with ~20% topping 30% ROI. This isn’t a contradiction; it’s two views of the same funnel—MIT measured the pilot pile (most stall), Deloitte measured the survivor program (most pay). That’s operations done right.

Momentum hasn’t cracked; it’s maturing. 78% of organizations already use AI in at least one function, and the winners are treating it like ops, not a stunt—shipping into real queues, tracking cycle time and cost-to-serve, and letting models earn trust shift by shift. Less sizzle, more fixtures. That’s how this market gets healthier—and how good AI tools prove their value.

Help Me Improve Failory

That's all of this edition.

Cheers,

Nico