Good Morning! Here's what I have for you in today's newsletter:
OpenAI adds Usage and Spend, letting teams track costs by API key
ComfyUI brings Qwen Image 3.0 Pro live with 4,500-token prompts
Databricks introduces Unity AI Gateway, to control every model and MCP tool
Stop repeating yourself to Claude Cowork with today's setup guide
4 new AI tools worth trying today
TRENDING NEWS
OpenAI added Usage and Spend dashboards that let developers track API usage and cost by individual API key, see which apps and workloads are actually driving spend, and set monthly limits at the organization or project level, including hard limits that cut off traffic once a threshold is reached.
Breaking down usage by API key means a team running multiple apps or workloads on one account can finally see exactly which one is responsible for a spike in cost, rather than only seeing one combined total.
Hard spend limits stop traffic automatically once a set threshold is reached, giving teams a real ceiling instead of just an alert that arrives after the overage already happened.
Limits can be set at either the organization or project level, letting a team apply broad guardrails company-wide or scope tighter budgets to specific projects.
Unpredictable API costs have been a recurring pain point for teams building on usage-based pricing, especially once multiple apps or workloads share the same account. Per-key visibility turns a vague monthly bill into an actual breakdown teams can act on, and hard limits remove the risk of a runaway process quietly running up charges overnight.
AI TOOL
ComfyUI brought Qwen Image 3.0 Pro live via Partner Nodes, supporting prompts up to 4,500 tokens, 4.5 times the previous generation, legible text down to roughly 10 pixels, 12 languages, 20 or more fonts, and support for formulas, diagrams, and nested UI mockups.
The 4,500-token prompt limit lets users describe far more detail and structure in a single request than the previous generation supported, useful for complex layouts rather than short, simple prompts.
Legible text at roughly 10 pixels solves a long-standing weak point in AI image generation, where small or dense text has historically come out garbled or unreadable.
Support for formulas, diagrams, and nested UI mockups pushes the model beyond typical creative image generation into more technical and structured visual outputs.
Small, legible text and structured layouts have been two of the more persistent gaps between AI image generation and genuinely usable design output. Qwen Image 3.0 Pro closes both at once, turning image generation into something that can realistically produce posters, slide decks, infographics, and product mockups rather than just standalone artistic images. Multi-image editing and support across 12 languages also broaden who this is actually useful for, beyond a narrow English-only creative use case.
AI TOOL
Databricks introduced Unity AI Gateway, giving admins one place to see usage and costs across every model, coding agent, and MCP tool in use, set shared or per-user budgets, control which MCP tools agents can access, and block sensitive PII before it ever reaches an external model.
A single dashboard consolidates usage and cost tracking across every model and MCP tool a team has connected, replacing what would otherwise be scattered visibility across separate integrations.
Budgets can be set as shared pools or scoped to individual users, giving admins flexibility in how spending limits get distributed across a team.
Input guardrails block sensitive PII before it reaches an external model, adding a real data-protection layer at the point where a request would otherwise leave the organization's control.
As teams connect more models, coding agents, and MCP tools, the number of separate places to monitor usage and enforce policy multiplies right along with it. Centralizing that visibility and control into one gateway addresses a real operational gap, especially the PII-blocking piece, which closes off a genuine risk of sensitive data leaving an organization through an external model call.
HOW TO AI
Every new Cowork session starts the same way. You re-explain your tone, re-paste your rules, remind it what you hate. The fix isn't a better prompt. Cowork reads a folder of plain text files before it says a word, and if you've never opened that folder, you've been talking to a stranger every time.
Set up your Cowork folder
Create a folder called Claude Cowork. Inside: about-me (your voice, taste, and what you hate sounding like), claude-output (Cowork's saves), templates (favorites to match format later). If Cowork won't trust the folder, put it directly in your Home directory, not inside Desktop or Documents.
Point Obsidian at the same folder
Cowork's files are plain markdown, unreadable in a default text editor. Download Obsidian free, click Open folder as vault, select your Claude Cowork folder. Nothing converts, it's just a nicer window onto the same files.
Edit here, Claude remembers there
Add a real line to about-me.md:
Always ask me three clarifying questions before starting any task,
never assume.Start a new Cowork task, and it asks clarifying questions unprompted, proof the file carried over.
Capture research the moment you find it
Obsidian Web Clipper, free and official, saves any page straight into your vault as clean markdown.
Build a skill just by talking to it
Repeated tasks are worth turning into a Skill:
Create a skill called "negotiation." Interview me about the kinds of
deals I negotiate regularly. Save it as a skill I can call with
/negotiation.Cowork interviews you, builds it, shows examples first. Test with /negotiation. Same pattern works for /weekly-report (pulls live data from connectors) and /client-brief (onboarding new clients).
P.S. You can access all the AI trainings, prompts and workflows if you upgrade.
Google DeepMind announced a research partnership with Fenris Creations to study how AI agents learn inside a living, persistent game universe, extending SIMA's 3D-world research into continual learning, deep memory, long-horizon planning, and multi-agent dynamics.
ImagineArt introduced Variable and Carousel nodes in its Workflow tool, letting anyone attach a variable node to add flexibility or pull any single frame and turn it into a full carousel.
Firecrawl introduced Developer Index, an index built specifically for coding agents that searches over 70 million primary sources including repos, docs, and issues, offering the highest recall of any coding-specific index to help agents ship correct, up-to-date code.
🚀 Okara: the world's largest launch library, browse the best launches on X and study why they worked before your own.
🔍 Is Agentic: free agent-readiness scoring tool, scans any public website across 100+ checks and returns prioritized fixes, by Vercel.
🎬 MiniMax Design: an agent-driven commercial content studio that takes a goal and autonomously plans and produces ads, e-commerce assets, and motion graphics.
📝 Notion Skills: turn any repetitive task into a reusable skill Notion Agent can run automatically, or export as a SKILL.md package for Claude Code, Codex, Cursor, Gemini, or Grok.
THAT’S IT FOR TODAY
Thanks for making it to the end! I put my heart into every email I send, I hope you are enjoying it. Let me know your thoughts so I can make the next one even better!
See you tomorrow :)
- Dr. Alvaro Cintas
✓ Full archive of premium guides with ready-to-use prompts
✓ Structured AI courses (step-by-step, start-to-finish)
✓ Every upcoming premium tutorial