AI Ate Uber’s Budget Before April Ended — AI Brief April 25
Today’s Context Window includes Google’s biggest Gemini overhaul yet, ComfyUI’s $500M creator economy bet, Andrew Yang making the case for UBI on Hard Fork, and ByteDance’s new 3D model that’s reshaping what content creation means.
Good morning, humans. Uber blew through its entire 2026 AI budget before May — and the tools doing it are writing 11% of the company's live backend code without a human in the loop. Meanwhile, Google just shipped six new Gemini features in one drop and a 3D model from ByteDance is making game assets and product mockups faster and cheaper to generate. It's a busy Saturday. Let's get into it.
📬 Before we dive in: The sharpest AI Brief tips come from readers who are actually in the weeds. If you spot a story worth covering — a quiet research drop, a product launch buried in a press release, a shift that isn't getting attention yet — share it in the community chat. The best tips make tomorrow's edition.
Uber burned its entire 2026 AI budget by April
What happened: Uber's CTO Praveen Neppalli Naga revealed the company has already exhausted its full-year AI budget — just four months in. The culprit: engineers adopted AI coding tools so aggressively that 11% of live backend code updates are now written entirely by AI agents, with about 1,800 code changes per week happening without direct human input. Nearly 70% of committed code is AI-generated. Monthly API costs per engineer are running $500–$2,000.
Why it matters: This is the first major public case of a Fortune 500 company blowing past its AI budget not because of a failed project — but because the tools worked too well. The ROI question just got a lot more complicated: if AI is writing your code but the cost is uncapped, what exactly did you save?
What everyone's saying: The "tokenmaxxing" trend is getting attention — companies like Meta rank employees on internal dashboards based on how many tokens they consume. Critics argue high token usage doesn't equal better outcomes, just higher bills. Uber is now testing OpenAI's Codex to diversify its AI stack and rein in costs.
My read between the lines: Uber ranked engineers on leaderboards for Claude Code usage — essentially gamifying their own budget crisis. The lesson isn't that AI is too expensive. It's that the cost structure of AI development is fundamentally different from traditional software: there's no upper bound on spending if the model is always ready to work. That's a new CFO problem that didn't exist two years ago.
Today's AI Brief is brought to you by MirrorMemory.ai — your AI-powered second brain.
Google's April Gemini Drop: six features, one big vision
What happened: Google shipped the 10th edition of its monthly "Gemini Drop," delivering six features at once: NotebookLM is now built directly into the Gemini app, Personal Intelligence expands globally (connecting Gmail, Photos, Drive, and Calendar to your AI assistant), a native Mac app launched, users can generate 3-minute AI music tracks with Lyria 3 Pro, 3D model and interactive chart generation landed in-app, and Google added tools to import your full ChatGPT or Claude conversation history into Gemini.
Why it matters: Google is collapsing all of its AI bets — Gemini, NotebookLM, Google Photos, Workspace — into a single surface. For anyone who felt like Google's AI was scattered across too many products, this is the consolidation moment. The fact that they're letting you import your ChatGPT history is a direct shot across OpenAI's bow.
What everyone's saying: The NotebookLM integration is getting the most enthusiasm from power users, who've long wished it was easier to access without switching apps. The Lyria 3 Pro music feature is being compared favorably to Suno and Udio — free, up to 3 minutes, higher fidelity. The "switching tools" for importing ChatGPT history are being read as a sign Google thinks it's finally competitive enough to ask for converts.
My read between the lines: The Personal Intelligence feature — where Gemini connects to your Gmail, Calendar, and Photos — is the most strategically important thing in this drop, and it's getting the least attention. Google has access to more personal data than any company on earth. If users actually opt in, Gemini becomes something no one else can replicate. The question is whether privacy fears or habit will win.
ComfyUI raises $30M at $500M valuation
What happened: ComfyUI — a node-based workflow tool that lets creators wire together AI image, video, and audio generation in granular steps — raised $30 million led by Craft Ventures at a $500 million valuation. Total funding is now $48 million. The startup started as an open-source project in 2023 and became so popular among professional creators that it evolved into a full company. The CEO's pitch: text-prompt AI tools only get you 60–80% of the way there; ComfyUI lets creators control the rest.
Why it matters: Midjourney and ChatGPT image gen are great for getting started fast. But professionals — game studios, filmmakers, product designers — need to control exactly what they're making. ComfyUI is the "Photoshop for AI" play: the tool that lets you do it precisely instead of hoping the model guesses right.
What everyone's saying: The funding signals that investors believe AI creative tools won't consolidate entirely around text prompts. The "node-based workflow" niche is familiar from tools like Blender and DaVinci Resolve — visual, composable, and beloved by people who actually produce things professionally. The creative professional market is being bet on as durable.
My read between the lines: A $500M valuation for a tool that is still largely open-source is the VC community betting that the real AI creative economy isn't the one where everyone types a prompt — it's the one where professionals build production pipelines. If that's right, ComfyUI is infrastructure. And infrastructure gets priced like infrastructure.
Andrew Yang makes the case for UBI on Hard Fork
What happened: On the latest episode of the NYT's Hard Fork podcast, Andrew Yang joined Kevin Roose and Casey Newton to argue that universal basic income is becoming an economic inevitability as AI displaces workers. Yang — who ran for president in 2020 on a UBI platform when the idea seemed fringe — is making the case that the displacement he warned about is visibly arriving: entry-level hiring is declining, AI tools are replacing cognitive tasks, and the political window for a policy response may be closing.
Why it matters: The UBI debate is no longer theoretical. Meta, Uber, Snap, and dozens of other companies are reporting record productivity while cutting headcount. When the companies doing the cutting are thriving — not struggling — that's when the policy question becomes urgent: where do the productivity gains go, and who decides?
What everyone's saying: Yang's 2020 campaign felt ahead of its time; now it's being revisited as prescient. The Hard Fork hosts pushed back on whether UBI is politically viable and whether the tech industry itself would support it. The episode also covered Tim Cook's Apple legacy — a reminder that AI disruption isn't just about startups, it's reshaping the companies that built the devices everyone uses.
My read between the lines: The uncomfortable reality Yang is circling: the people most likely to advocate for UBI are the engineers building the AI — and they're also the ones whose jobs are most insulated from it, at least for now. The case for UBI isn't going to come from Silicon Valley's conscience. It'll come when the economic data becomes impossible to spin as "transition."
ByteDance launches Seed3D 2.0: sharper geometry, better textures
What happened: ByteDance released Seed3D 2.0, a new 3D generation foundation model that achieves state-of-the-art performance in both geometry accuracy and texture quality. Unlike text-to-image models, Seed3D generates full three-dimensional objects — with physically accurate materials and complex structures — for use in games, films, product visualization, and virtual environments. The API is live via Volcano Engine, ByteDance's cloud platform.
Why it matters: 3D asset creation is one of the most expensive and time-consuming parts of game development, film production, and e-commerce. A model that generates state-of-the-art 3D objects from inputs collapses months of work into minutes. This isn't a demo — it's a cost structure disruption for entire creative industries.
What everyone's saying: The quality jump in Physically Based Rendering (PBR) materials — meaning objects look realistic under real lighting conditions — is getting the most attention from game developers and VFX artists. Seed3D 2.0 is being discussed alongside 3D generation advances from Stability AI and OpenAI's internal research as evidence the space is maturing fast.
My read between the lines: ByteDance keeps shipping foundation models across modalities — video (Kling), music, and now 3D — and quietly deploying them via Volcano Engine, which is their AWS play. TikTok is the brand everyone talks about. Volcano Engine is where ByteDance is building the infrastructure business. Seed3D 2.0 is another tile in that mosaic.
That's your AI Brief for Saturday, April 25. Spotted something we missed, or have a take on today's stories? Join the conversation in the Artificially Intimidating community chat — the best insights always come from readers.
—Artificially Intimidating