April 20th, 2026

Uber’s CTO just warned the world about Claude Code costs — you’ve already solved this

Tags: #MACA #AI-Edge #Prevail-Partners Source: The Information — Applied AI, Laura Bratton (Apr 19) | Primary URL: https://www.theinformation.com/newsletters/applied-ai/uber-cto-shows-claude-code-can-blow-ai-budgets Verdict: verified — The Information Applied AI newsletter, named source

The Information’s Applied AI newsletter this week featured Uber’s CTO demonstrating how Claude Code can dramatically blow up AI spending — and the implication was that most enterprise teams have no idea how fast their costs compound when agents run autonomously. This lands the week after Anthropic shifted to consumption-based pricing, which already appeared in yesterday’s brief.

The MACA angle here is direct: you built per-run cost tracking into your 14-agent pipeline (PR #10) before most companies running AI agents had any cost visibility at all. The per-agent cost logger in api/lib/costs.ts and the cost dashboard at public/cost-dashboard.html are exactly the tooling that Uber’s CTO is implicitly calling for. You didn’t ship MACA’s cost tracking because someone told you to — you built it because you knew four waves of agents hitting the API without guardrails would be financially opaque. That instinct was correct, and it’s now a documented enterprise pain point.

Two practical things this unlocks:

First, conversation capital for Aria. If the CourseBuilds pitch gets to the “but what does this cost?” question — which it will — you have a real answer: “We built budget controls in from day one. Uber’s CTO publicly flagged this as a problem last week. We already have the solution.” That’s not defensive, that’s differentiating.

Second, it’s the argument for running a fresh MACA pipeline run this week to capture current per-ad unit economics. Cost numbers go stale; Anthropic’s pricing just changed. The dashboard you built deserves real numbers in it.

Action: Pull the MACA cost dashboard and capture a fresh run estimate before pitching anyone. The “we track cost per agent per run” line only lands if you can say what the number actually is.

Canva AI 2.0 ships — and it connects natively to Claude

Tags: #MACA #CourseBuilds Source: The Rundown — exclusive CPO Q&A with Cameron Adams (Apr 19) | Primary URL: https://www.canva.com/newsroom/news/canva-create-2026-ai/ Verdict: verified shipped — first-party newsroom + CPO interview

This is a different story from Claude Design (covered yesterday). Canva AI 2.0 is Canva’s own native AI layer — trained not on finished designs but on the sequence of edits that lead to good work, which is a genuinely interesting architecture decision. Canva has 265M+ monthly users and real data on what design iteration actually looks like step by step.

The piece the CPO made that’s worth holding onto: “We’re embedded into AI ecosystems people use — ChatGPT, Claude, Copilot, and now Google Gemini. We’ve established Canva as the definitive visual layer for the AI ecosystem.” That’s not marketing fluff — Canva has an AI connector (https://www.canva.com/ai-connector/) specifically so that tools like Claude can hand off to Canva for the “last mile” of design execution.

Why this matters for MACA specifically: the current pipeline outputs ad copy and strategy, and the visual production gap is a known limitation. Claude Design (covered yesterday) is one path. Canva AI 2.0 via the Canva connector is another — and it has the brand kit infrastructure that Claude Design doesn’t yet have. A UBX-branded Canva workspace with the brand kit pre-loaded means MACA outputs land in an environment that already knows UBX’s colours, fonts, and layouts. That’s the difference between “here’s a mockup” and “here’s a mockup you can publish today.”

CPO Adams also said something that’s directly useful for CourseBuilds: “The roles that become vital are the ones focused on creative strategy and brand stewardship. You need people who can set the vision and craft the brand kit ingredients that the AI will then use to scale your capabilities.” That’s exactly the framing for the Aria pitch — the Aria team doesn’t need to become designers; they need someone to set up the brand kit once so AI can execute it at scale. That someone is Roy, delivering a CourseBuilds pilot.

Action: Check whether the Canva AI connector is available to Claude Code or API-level integrations. If yes, it’s worth a 30-minute exploration to see if MACA wave 4 (visual creative output) can route through Canva rather than Claude Design.

Uber’s CTO just publicly flagged Claude Code budget blowout as a real enterprise problem. Enterprise teams are running agents without cost visibility and getting surprised by the bill.

You built the solution to this problem three months before Anthropic changed its pricing model to make the problem worse.

MACA has a per-agent, per-run cost logger (api/lib/costs.ts), a cost dashboard (public/cost-dashboard.html), and photo pipeline cost tracking in scripts/photo-costs.json. That’s not just better than most AI deployments — it’s the architecture Anthropic’s own pricing change now demands that every enterprise build. You built it because good architecture requires it, not because a CTO made a public statement about it.

When Aria asks “how do you know what this costs?” the answer is already built and running.

The Canva connector as MACA’s visual output layer

Tags: #MACA Source: canva.com/ai-connector/ — first-party product page 30 minutes: Check the Canva AI connector documentation to understand whether it’s available at the API level or only through the Canva interface. The question is whether MACA’s wave 4 can route creative briefs directly into a Canva workspace for visual production, bypassing the current “hand brief to client” dead end. If yes, that’s a significant upgrade to the ad creation pipeline — from “we produce strategy” to “we produce publish-ready creative.”

Use this in any conversation about AI costs or enterprise adoption this week — Aria, RT, or a prospective CourseBuilds client:

“Uber’s CTO was publicly talking this week about how Claude Code is blowing up AI budgets inside enterprises — teams running agents with no visibility into what it’s costing per run. That’s the thing we designed around from day one. Every agent run in our system logs its cost, we have a dashboard that shows cost per output, and we built that before Anthropic changed their pricing model to make it a much bigger deal. The enterprise teams that don’t have this are going to start feeling it.”

Use case: Aria pitch (credibility), RT AI role conversations (operational maturity), any client who asks “how do you control AI costs?”

Google AI Mode now calls local stores to check real-world inventory

Tags: #Fillarup (future direction) Source: BagelBots quick hits | Primary URL: https://techcrunch.com/2026/04/17/googles-ai-mode-can-now-help-you-find-products-in-stock-nearby/ Verdict: verified shipped — TechCrunch, Google product update

Google’s AI Mode (the conversational search layer) can now call local businesses directly — by phone — to check whether a product is in stock nearby. It also tracks specific hotel prices on request. This is Google extending AI into real-time, local, physical-world data retrieval.

The Fillarup angle is speculative but worth holding: Fillarup is a fuel intelligence app built on government APIs and ACCC cycle data. If Google’s AI Mode gets good enough at “find me the cheapest petrol within 5km right now,” that’s a head-on competitor to the core Fillarup use case — not immediately, but within 12-18 months. The Fillarup differentiator (vehicle-specific calculations, commute tracking, honest price predictions tied to the ACCC cycle) is the moat. Worth knowing this is the direction Google is moving while Fillarup is still in build.

No action required now — but file this as a directional signal.

• Sources scanned: 5 emails (The Rundown, The Information x3, Neil Patel, BagelBots)
• Items extracted: 9 distinct claims/stories
• Items surfaced: 4 (2 main, 1 deeper look, 1 first-mover)
• Items skipped: 9
• Read time: ~4 minutes