The cost of AI reasoning just dropped 95%
Plus: Google's Nano Banana Pro launch, Relay adds native tables, and why smart agencies are self-hosting their agents.
Welcome to issue #87 of FutureBrief. Three times a week I share practical insights on AI & automation trends, tools, and tutorials for business leaders. If you need support on your technological journey, join our community and get access to group chat, Q&As, workshops, and templates.
Ninjabot, a ready-to-deploy AI agent that responds to inquiries in 10-60 seconds (not hours), reaches out to relevant contacts on social media 24/7 (no spam), and warms up your database via SMS/email (including follow-ups).
🔮 Today’s insights
OpenAI fast-tracks GPT-5.2 to fight Gemini
Reports indicate OpenAI has declared an internal Code Red and may release GPT-5.2 as early as today, December 9. This defensive move aims to counter Google’s Gemini 3, which recently overtook ChatGPT in key reasoning benchmarks. If you rely on ChatGPT for complex reasoning or coding, expect a major performance jump this week. The competition is forcing rapid improvements that benefit your business immediately.
Google launches Nano Banana Pro for Workspace
Google rolled out Nano Banana Pro, a new image generation model integrated directly into Workspace apps. It features Generative UI capabilities that can create dynamic visual layouts and editable assets inside Slides and Docs. Marketing teams can now generate and edit professional assets inside their existing workflow without needing external tools like Midjourney or Canva.
Acuity launches anti-manipulation scoring
A new tool called Acuity scores news content for emotional coercion and tribal engineering rather than just factual accuracy. It identifies when text is structurally designed to bypass critical thinking. It will be a valuable tool for fact-checking and research.
Relay launches native Tables
Relay introduced Tables, allowing users to store, query, and manage data directly within their automation workflows. This removes the need to constantly ping Google Sheets or Airtable for simple state management. You can now build lightweight CRMs or approval trackers entirely inside your automation platform. It reduces API usage, speeds up workflows, and keeps your data close to the action.
💡 The commoditization of reasoning is finally here
I am noticing a massive shift in the economics of automation this week. DeepSeek V3.2 launched with pricing at $0.28 per million input tokens. That is roughly 95% cheaper than comparable frontier models like GPT-5.
For the last year, we’ve had a couple of clients who hesitated to build complex agentic workflows because the math did not work. An agent that thinks before it acts consumes thousands of tokens per task. At OpenAI prices, running that loop for customer support could cost thousands of dollars a month for a busy team.
DeepSeek changes the equation entirely. I can imagine businesses switching their backend infrastructure to DeepSeek fast. Not for the cost savings, but to run verbose loops. You can let the AI think for 100 steps to solve a hard problem without destroying your margins.
This is the Linux moment for AI agents. Intelligence is becoming a cheap commodity. The competitive advantage for your business is no longer about access to the smartest model. It is about how well you orchestrate that cheap intelligence to do real work.
My advice: Stop worrying about token costs. Start building workflows that let the AI think as much as it needs to.
🔧 Tools & Resources
DeepSeek V3.2: Open-source reasoning model, best for running high-volume, complex agentic workflows at 95% lower cost than GPT-5.
Relay.app Tables: Native database for workflows, best for storing and querying data directly inside your automation steps without needing Google Sheets.
Nano Banana Pro: Google’s Gen-UI model, best for creating editable marketing assets and UI mockups directly inside Google Workspace.
Softr: No-code platform to build client portals and internal tools from Airtable or Google Sheets. Best for businesses needing custom dashboards without hiring developers.
💬 Community Q&A
Q: “DeepSeek sounds great and cheap, but is it safe for client data?”
A: Because DeepSeek is open-source, you can self-host it on your own private cloud or servers. This is arguably safer than using OpenAI because the data never leaves your infrastructure. However, this is only a viable option if you have developers on your team or an AI automation agency that can help you set up and manage the hosting environment.
🎓 Next workshop
Build Your AI Research Assistant in 60 Minutes
Replace 10 hours of weekly research with an AI system for competitor analysis and market trends. Premium members build it live with Yuri.
Forward this to a colleague who’s wrestling with manual processes. They’ll thank you.
What’s your take on today’s topics? Did you like it, or is there something I missed?
Building modern tech for SMBs? Reach 20,000+ decision-makers who are actively implementing AI, automation, and no-code tools. Become a sponsor
Incredible take on the cost collapse here. The "Linux moment" framing is spot-on, but I think the bigger unlock isnt just cheap tokens, its that verbose reasoning loops can now run in production without blowing budgets. Teams have been timid about multi-step agentic workflows becuase every loop iteration added cost. Now that constraint disolves, and we'll see agents doing way more upfront thinking.