• Agent Pulse
  • Posts
  • #35: Google Outsmarts Olympiad, Mistral Finds Its Voice, and Netflix Rewrites with GenA

#35: Google Outsmarts Olympiad, Mistral Finds Its Voice, and Netflix Rewrites with GenA

AgentPulse: Your Weekly Dose of AI Agents News.

Author

O P
July 23, 2025

In partnership with

Welcome back, AI Agent Enthusiast!

In today’s Agent Pulse:

  • 📢 Top Headlines

  • ⚔️ Agent Arena

  • ✨ Featured Agents

  • 📡 Agent Signals

  • 🎓 Free courses

  • 📚 Must-Read Papers

  • 🗺️ Landscape Map

You read, we listen. Got feedback? Just hit reply - we’d love to hear from you.

Enjoying Agent Pulse?
Go ad-free and support our work for just $5/month. Upgrade to Premium →

Meet your new assistant (who happens to be AI).

Skej is your new scheduling assistant. Whether it’s a coffee intro, a client check-in, or a last-minute reschedule, Skej is on it. Just CC Skej on your emails, and it takes care of everything:

  • Customize your assistant name, email, and personality

  • Easily manages time zones and locales

  • Works with Google, Outlook, Zoom, Slack, and Teams

  • Skej works 24/7 in over 100 languages

No apps to download or new tools to learn. You talk to Skej just like a real assistant, and Skej just… works! It’s like having a super-organized co-worker with you all day.

📢 TOP Headlines

Google DeepMind just broke another frontier: an enhanced version of Gemini Deep Think scored 35/42 on the 2025 International Mathematical Olympiad (IMO), earning an official gold-medal rating, the first time such recognition has been granted to an AI system

What Changed This Year

  • Unlike last year's DeepMind models, Gemini solved five out of six IMO problems directly from natural language, within the same 4.5-hour time frame students use

  • It uses Deep Think mode, which deploys parallelized reasoning and reinforcement learning, trained on theorem-proving and high-quality math solutions

  • Notably, IMO judges officially graded the output, validating the model’s solutions as rigorous proofs, not just plausible answers

OpenAI Goes Gold Too

OpenAI also announced a gold-tier performance, matching Gemini’s 35/42, though it self-reported the result rather than undergoing the official grading process, triggering debate about credibility

Why It's a Big Deal for Agent Builders

  • This isn’t benchmark performance, it’s certified, domain-level reasoning under real-world constraints. Agents now have validated capabilities at the highest human reasoning levels.

  • Natural-language reasoning across steps signifies that agents can autonomously parse, plan, prove, and respond, in competition-quality depth.

  • With official grading, we might finally start trusting agent outputs for high-stakes context, creating opportunities in areas like legal reasoning, academic publishing, and scientific discovery.

Takeaways

  • Agents are now certified collaborators, not just tools, they can meet human-level standards in rigorous reasoning environments.

  • The gap between “reasoning LLMs” and “reasoning agents” is collapsing, agents are no longer fuzzy assistants, but trusted arbiters of correctness.

  • What comes next is multimodal agentic reasoning, applying the same rigor in areas like physics problem solving, data analysis, and scientific workflows.

Mistral just dropped Voxtral, a breakthrough open-source audio model family that redefines what's possible in voice AI, offering both scale and semantic understanding with production-ready utility

What It Does

  • Voxtral Small (24B) and Voxtral Mini (3B) support 30–40 minutes of continuous audio transcription plus Q&A and multi-language summaries, no chains of tools needed

  • Underperforms none, outperforming Whisper large-v3, GPT‑4o mini Transcribe, Gemini 2.5 Flash, and even ElevenLabs Scribe, across multiple languages and benchmark tasks

  • Built-in function calling on voice allows it to trigger workflows directly from speech, “true speech-to-action” without glue code

Why It Matters

  • Free + open + business-grade: Voxtral is open-source under Apache 2.0 and available for self-hosting or via API at ~$0.001/min, about half the cost of Whisper-based APIs

  • Edge-ready option: The 3B Mini variant is optimized for local deployment, ideal for embedded systems, IoT, or on-device assistants

  • Enterprise-grade flexibility: Mistral also offers private GPU deployment, domain-specific fine-tuning, speaker/audio segmentation, emotion recognition, and multi-speaker diarization support for high-security environments

Takeaways

  • If you're building agentic voice workflows, Voxtral lets you unify transcription, context understanding, and action in a single model.

  • Its hybrid reasoning, audio + language, signals a new class of voice agent: high-context, multilingual, function-enabled.

  • As an open model, it invites customization and experimentation, a contrast to closed audio stacks from big providers.

Voxtral crushes the precedent, open-source voice agents can now be fast, smart, cheap, and deployable at scale. If your agent roadmap includes spoken interaction, this is your new baseline.

Netflix has taken a bold step: generative AI now appears in final footage of its original series The Eternauta. According to Co‑CEO Ted Sarandos, a complex scene showing a building collapse in Buenos Aires was built using GenAI, delivered 10× faster and at a fraction of cost compared to traditional VFX techniques. At 96% on Rotten Tomatoes, the show debunks any notion that AI-driven production lowers quality

But Netflix isn’t stopping there, this marks the beginning of a broader integration:

  • GenAI is now fueling personalized search, letting users describe what they want to watch in natural language, whether “something funny and upbeat” or “not too scary, but a bit funny”

  • The company plans to introduce AI-generated mid-roll and pause ads by 2026 on its ad-supported tier, blending product placement and interactive elements contextually within shows

Why This Matters for Agent Builders

  • AI as Co-Creator, Not Just Tool: Netflix is moving GenAI into the core production pipeline, not just preview or ideation. Agents are now billable in visual output.

  • Watching Along, Agent-Side: When the same audience witnessing storytelling sees contextual AI ads and conversational discovery features, trust and expectation shift, your agents must feel integrated, not tacked on.

  • Data-Savvy Creativity: As Netflix layers AI into search, content, and advertising, it amplifies a unified agentic loop: recommend → create → monetize, all while personalizing per user.

Takaways

Agents in 2025 are not assistants, they’re embedded creators. The path Netflix is forging shows how AI can seamlessly become part of both content and context. As you design agents, whether for media, commerce, or workplace automation, ask yourself: is your agent just functional, or capable of creating on-screen value in trusted and transactable ways?

AI You’ll Actually Understand

Cut through the noise. The AI Report makes AI clear, practical, and useful—without needing a technical background.

Join 400,000+ professionals mastering AI in minutes a day.

Stay informed. Stay ahead.

No fluff—just results.

AArena: All-in-One AI Workspace

Get Sh*t Done with AI.
Bring your task. Get best results.

TOP 5 AI (July 22):

  1. Llama 3.3 70B Instruct

  2. ChatGPT-4o

  3. Llama 3.3 70B Instruct Turbo

  4. Grok 3

  5. Grok 3 Fast

✨ Featured Agents

Oraczen helps enterprises rewire their workflows with Agentic Systems powered by the Zen Platform. Their industry-specific solutions go beyond automation - they think, adapt, and deliver measurable outcomes. Built for intelligence and flexibility, Oraczen products enable organizations to accelerate innovation, enhance decision-making, and achieve real, transformative business results

  • SDKs for Enterprise: Readers may not realize we offer SDKs to build their own agentic systems.

  • Plug-and-Play AI Products: In addition to the SDKs, they can deploy prebuilt agentic systems like our Spend Analyzer, Conversational AI Assistant, and Invoice Processor—all tailored for fast enterprise adoption.

👉 Explore

📡 Agent Signals

  • OpenAI drops ChatGPT Agent – A single “agent mode” now blends ChatGPT, Deep Research and Operator so the model can plan, browse and code its way through multi-step tasks like booking travel or building slide decks; Pro users get 400 runs/month, Plus/Team get 40.

  • Amazon launches AgentCore + AI Agents Marketplace – A new toolkit and storefront with 900+ ready-made agent plug-ins (Stripe, IBM, Anthropic) backed by a $100 m innovation fund to help enterprises spin up secure, sandboxed agents at scale.

  • Pipe debuts four fintech agents – Fraud, compliance, capital-payments and customer-engagement agents already compress “weeks of ops into near-instant workflows” as the embedded-finance platform gears up for global expansion.

  • Composio, officially known as Sampark Inc., an AI-driven SaaS startup focused on enterprise workflow automation, has announced it raised $25 million in a Series A funding round led by Lightspeed Venture Partners.'

  • Blaxel nabs $7.3 m seed – First Round Capital leads the round for the “AWS-for-agents” cloud that spins up sandboxed VMs in seconds and already handles >1 bn agent-seconds a month for customers in 16 regions.

  • Magentic closes $5.5 m – Sequoia backs the procurement-focused agent startup that spots tariff-driven savings and auto-negotiates supplier deals.

  • Perplexity ships Comet browser (beta) – Chromium-based browser with an embedded agent that can book flights, restaurants and surface deals while you keep tab-surfing.

  • Silverback AI Chatbot adds memory-rich Agents – Moves beyond scripted flows to context-aware agents that remember past chats, update CRMs and hand off to humans only when needed.

  • CYE launches Hyver AI agent – Cyber-exposure agent ingests 1 k+ enterprise tools, auto-maps attack graphs and triggers templated remediation workflows.

  • VIREAS energy assistant goes live – Croatian regional energy agency’s new chatbot walks homeowners through renovation subsidies, heat-pump sizing and solar grants in plain language.

  • 97.5 % of dev teams now use AI in the SDLC – Techreviewer’s 2025 survey of 20+ countries shows AI code-gen (72 %), doc & review (67 %) and auto-testing (56 %) as the top use-cases; 82 % report ≥20 % productivity lift.

🎓 FREE Courses

📚 Must Read Papers

  1. KPMG: AI Quarterly Pulse Survey: Q2 2025 (Doc)

  2. Stanford University: Future of Work with AI Agents (Doc)

  3. Google: Guide for using AI at work (Doc)

  4. Google: An Introduction to AI Agent Security (Doc)

  5. Thomson Reuters: Agentic AI 101 (Doc)

  6. OpenAI: A Practical Guide to building Agents (Doc)

  7. BCG: AI at Work (Doc)

  8. ServiceNow: Enterprise AI Maturity Index 2025 (Doc)

  9. IBM: Agentic AI in Financial Services (Doc)

  10. Capgemini: Rise of Agentic AI (Doc)

🗺️ AI Agents Landscape Map

Discover the evolving AI agents ecosystem. We’re mapping thousands of vendors to help you quickly find the right solution for your needs.

THANK YOU

Visit website 

Follow us on Linkedin, X

I appreciate your time.

OP & Team

How'd we do?

Login or Subscribe to participate in polls.

Reach 23,000+ Readers:

Newsletter is read by VCs, founders, engineers, managers and tech professionals.

Reply

or to participate.