Hey - It’s Nico.

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

If you only have one, here are the 3 most important things:

• Huxe, an AI audio app that turned your inbox into a personal daily briefing, has shut down — learn why below

• How to Avoid AI Code Slop

• Tokenmaxxing is the new obsession with measuring AI productivity — learn why it matters below

A huge thanks to today’s sponsor, SimpleClosure. When startups hit the wall, close cleanly and move forward with expert-guided support.

When It’s Time to Close, Do It Properly ^AD

Across the startup landscape, founders are facing tougher markets, crowded categories, and harder paths to sustainable growth.

When shutdown becomes the right next step, the work is not over.

SimpleClosure helps founders bring structure to the process, from state filings and investor communications to final distributions, compliance, and selling remaining assets. So you can close cleanly, responsibly, and with a clearer path through.

👉 Simplify Your Shutdown

This Week In Startups

🔗 Resources

How to Avoid AI Code Slop

In B2B customer support, AI is a Copilot, Not a Replacement

The Founder's Black Box

Anthropic and OpenAI have finally found PMF

📰 News

Meta launches Instagram, Facebook, and WhatsApp subscriptions

YouTube adds new podcast features, including an AI recommendation tool

Pope Leo uses first major papal text to warn about dangers of AI

Robinhood now lets your AI agents trade stocks

💸 Fundraising

AI coding startup Cognition raises $1B

Mercury, a fintech banking platform for startups, raises $200M Series D funding

Moment, an AI operating system for investment management, raises $78M Series C funding

Modal, a serverless cloud platform for AI infrastructure, raises $355M Series C funding

Fail(St)ory

Personalized Podcasts

Huxe was an AI audio app that turned your inbox, calendar, and interests into a personal daily briefing.

You opened it in the morning, pressed play, and got something closer to a podcast about your day.

It came from the same world as NotebookLM’s viral Audio Overviews. Ex-Googlers built it. Good investors backed it.

Then, on May 21, 2026, Huxe announced it was winding down, in the same month Spotify and Amazon pushed deeper into AI-generated personal audio.

What Was Huxe:

Huxe was a consumer AI startup in the Bay Area built around one bet: people do not want to read everything.

They want the important stuff spoken to them while they drive, walk, cook, work out, or avoid another screen.

The team came from Google’s NotebookLM project, where they had already seen AI-generated audio take off.

At first, Huxe looked more like a work tool. The early idea was to connect systems like Salesforce and let teams chat with company data.

Then the team shifted toward consumers. The bigger opportunity seemed to be audio.

The core product was a personalized daily briefing. Users connected email, calendar, and interests, then got a morning audio summary covering meetings, important emails, news, and topics they cared about.

The app also had personalized audio feeds, AI-generated DeepCasts built from prompts, live AI radio-style stations around topics like tech or sports, and interactive hosts users could interrupt to ask questions or change direction mid-conversation.

It was pretty ambitious. Huxe touched productivity, news, learning, search, and voice assistants all at once.

However, at the same time, Google, Spotify, and Amazon were pushing into the same category with products tied directly into platforms people already used every day.

Huxe raised $4.6 million in September 2025. Eight months later, it shut down.

The Numbers:

• 🗓️ Founded: late 2024, after the team left Google

• 💰 Funding: $4.6M

• 📱 Public launch: June 5, 2025

• 🛑 Shutdown announced: May 21, 2026

• 📉 Public launch to shutdown: about 11.5 months

Reasons for Failure: 

• The product was turning into a platform feature: A personalized AI briefing fits naturally inside Google, Spotify, Amazon, Apple, or Microsoft. Google had NotebookLM and Daily Listen. Spotify announced Personal Podcasts the same week Huxe wound down. Amazon had Alexa Podcasts. Huxe had to convince users to download a new app, while the platforms could place similar features inside products people already used every day.

• The trust barrier was high: Huxe worked best when users connected email, calendar, and personal context. That is sensitive territory for a young startup. Many users hesitate before giving inbox access to a product they discovered a few weeks ago. Big platforms already had the accounts, the data, and years of user familiarity behind them

• The product needed near-perfect reliability: The whole pitch was saving users time. That breaks fast if people still feel the need to double-check their inbox afterward. Some users liked the summaries but still worried important emails could get missed. Once that doubt appears, the product shifts from essential tool to optional layer.

• Audio has very little margin for error: Users tolerate mediocre text more than mediocre audio. You can skim weak writing in seconds. Audio forces you to sit through every awkward pause, robotic sentence, and bad transition. Small quality issues feel much bigger in a product people are supposed to use every morning.

Why It Matters: 

• In consumer AI, the biggest risk is often becoming a feature inside a larger platform.

• Audio products have a brutal quality bar because users feel every second of bad output.

Trend

Burning Tokens

This week, Uber’s COO said AI token spending is getting harder to justify. A few days earlier, OpenAI offered YC startups up to $2 million in API credits and encouraged founders to go “tokenmaxxing.”

That’s where the market is right now. One side is subsidizing massive AI consumption. The other is asking whether all this compute actually produces better products.

The interesting thing is how quickly tokenmaxxing escaped startup Twitter and became corporate behavior. Internal leaderboards. Usage quotas. Employees running agents for no reason just to look AI-active. 

A productivity tool quietly turned into a status game.

Why it Matters

• AI spend is becoming a hiring decision. Founders increasingly compare token costs against salaries, not software budgets. Uber even said it’s hiring fewer people while trying to balance AI spend against headcount.

• AI adoption and AI productivity aren’t the same thing. Companies like Meta and Amazon reportedly tracked token usage aggressively. Others, like Indeed, focused on outcomes instead.

• Tokens are becoming a real cost center. AI usage scales unpredictably through agents and automation loops. Companies are starting to treat tokens more like cloud infrastructure than SaaS seats.

Tokenmaxxing

Tokenmaxxing is what happens when companies start treating AI usage as proof of productivity.

Executives want employees using AI more aggressively. But AI adoption is hard to measure. Most workers say they use ChatGPT or Claude occasionally. Few actually change how they work around it.

So companies started looking for a measurable signal.

Tokens became the obvious choice. Every prompt, response, workflow, or agent loop consumes them. More tokens usually means more AI usage. That made token consumption an easy internal metric.

Then the dashboards appeared.

Meta built an internal leaderboard called “Claudeonomics” tracking AI usage across roughly 85,000 employees. Microsoft created internal token leaderboards as well. Amazon managers monitored AI adoption rates and usage dashboards internally. 

YC encouraged founders to think about “tokenmaxxing over headcountmaxxing.” OpenAI reinforced the trend by offering YC startups up to $2 million in API credits. Small startups started viewing tokens as a substitute for hiring.

Nectir reportedly pushed engineers to spend hundreds or even thousands of dollars per month on AI tools because leadership believed aggressive usage would force employees to discover better workflows. One founder described AI agents as becoming “an army of coders.”

Why it Broke

The problem with tokenmaxxing is simple: activity is easy to measure. Value isn’t.

Once token usage became visible inside companies, employees started optimizing for the metric itself.

According to reports, some Meta employees left AI agents running for hours just to inflate usage numbers. One user consumed 281 billion tokens in a month.

Amazon employees used internal AI tools for unnecessary tasks because they believed managers were paying attention to adoption dashboards

Then the financial reality showed up.

Uber burned through its annual Claude Code budget just four months into the year. Leadership started asking the question: does more token consumption actually create proportional business value?

Some people still believe tokenmaxxing is a useful signal, at least before employees start gaming it. Their argument: strong engineers using AI seriously naturally burn more tokens because they’re testing more ideas, running more workflows, and shipping faster. 

Others think the metric corrupts itself the moment companies start tracking it publicly. Employees stop optimizing for output and start optimizing for visibility instead. 

The real lesson is probably that token usage works as an early adoption push, but breaks once consumption becomes the goal rather than the byproduct of useful work.

Help Me Improve Failory

That's all for today’s edition.

Cheers,

Nico