Proč AI agenti potřebují více než jeden model: od volby mezi frontierovými a otevřenými modely po optimalizaci nákladů, rychlosti a kontroly nad citlivými daty. Praktický příklad z Glean a Walda.

Jak funguje systém modelů pro AI agenty v podnicích

Proč AI agenti potřebují více než jeden model

AI agents are built with many models, each bringing different strengths to the work. This explainer shows how powerful models can plan complex workflows while specialized models search enterprise data, call tools, review code, and validate results efficiently.

See how frontier and open models work together in production applications, intelligently matching each task to the right model to balance accuracy, efficiency, customization, and control.

Why AI Agents Need More Than One Model

Praktický návod, jak spravovat kredity Perplexity Computer ve firmě: co jsou Computer kredity, kdy se nečerpají (standardní vyhledávání) a jak fungují 2 vrstvy (měsíční seat kredity + sdílený fond).

Kredity Perplexity Computer v organizaci: jak je řídit a omezit náklady

Jak spravovat kredity Perplexity Computer v organizaci

Perplexity Computer credits are how organizations pay for agentic work — and how admins keep spend predictable. Standard Perplexity searches never consume credits; they're included in your subscription. Credits apply only to Computer: building dashboards and websites, running deep analysis across files and systems, and keeping recurring workflows alive for days, weeks, or months.

There are two layers to understand. Every seat comes with monthly credits — 500 per Enterprise Pro seat and 15,000 per Enterprise Max seat. Alongside that sits a shared organization credit pool that admins fund and govern for additional usage. Members spend their included credits first, then draw from the pool up to the limit an admin sets.

You'll learn how to:

Tell the difference between free Perplexity search and credit-consuming Computer work

Understand seat-level monthly credits for Enterprise Pro and Enterprise Max

Add credits to your organization pool from Organization settings

Set default caps by plan type and adjust limits for specific users

Configure auto-reload thresholds and monthly reload limits

Monitor per-user balances, pool balance, and current usage in one place

Seat credits reset each billing cycle and don't roll over. If credits run out, active work pauses and new credit-consuming tasks are blocked — then paused tasks resume automatically once credits are available. The goal is flexibility for teams and clear control for admins.

This is episode 6 of Perplexity Academy. Subscribe for more tutorials on Computer, Connectors, Skills, and enterprise administration.

🔗 Perplexity Enterprise: https://www.perplexity.ai/enterprise
📚 Perplexity Academy Playlist: https://www.youtube.com/playlist?list=PLKwRkjCH760NNEPqk75Ds6FfIMDmFwHIK

#PerplexityComputer #PerplexityAcademy #PerplexityAI #PerplexityEnterprise #ComputerCredits #AIAgent #AIWorkflow #EnterpriseAI #ITAdmin

Managing Computer credits | Perplexity Academy

Seznamte se s Birding Pal – digitálním parťákem pro birding, který reaguje na ptačí zvuky i slovní popisy. Ukázkově odhadne druhy, pracuje s nejistotou a vede přirozený dialog pro lepší pozorování.

Birding Pal: konverzační pomocník pro určování ptáků v terénu

Meet Birding Pal: nový digitální parťák pro pozorování ptáků

Your new birdwatching buddy is here. 
Make your own: https://github.com/openai/birdingpal

Meet Birding Pal

NVIDIA Alpamayo 2 Super je otevřený „reasoning“ model nové generace pro autonomní vozidla a robotaxi. Obsahuje 34 miliard parametrů, 360° vnímání okolí a propojuje strategické i provozní rozhodování v jednom systému.

NVIDIA Alpamayo 2 Super: otevřený reasoning model pro robotaxi a AV bezpečné rozhodování

NVIDIA Alpamayo 2 Super: otevřený model nové generace pro robotaxi a autonomní vozidla

NVIDIA Alpamayo 2 Super is here — the frontier open reasoning model for autonomous vehicles and robotaxis, now available for commercial use.

With 34B parameters and full 360° surround coverage, Alpamayo 2 Super handles trajectories, reasoning traces, meta-actions, and auto-labels in one model — streamlining what would've taken several models. 

It's open source under the OpenMDW-1.1 license — covering fine-tuning, derivative models, and commercial redistribution — giving AV developers a direct path from adaptation to deployment.

We built it, so you can build on it. 

📥 Download: https://huggingface.co/nvidia/Alpamayo2-Super
📝 Corp blog: https://blogs.nvidia.com/blog/alpamayo-2-super-open-model-now-available
➡️ Learn more: https://www.nvidia.com/en-us/solutions/autonomous-vehicles/alpamayo/

NVIDIA Alpamayo 2 Super: The Frontier Open Model for Robotaxis and Autonomous Vehicle

Jak NeuroDX AI využívá foundation modely, GPU NVIDIA H200 a podporu NVIDIA Inception k škálování analýzy EEG. V článku: dekódování „jazyka mozku“, nízká latence a přechod od občasného měření k akční zdravotní inteligenci.

NeuroDX AI & NVIDIA Inception: AI analýza EEG v reálném čase

Neurologické poznatky v reálném čase: Jak NeuroDX AI škáluje analýzu EEG s NVIDIA Inception

Traditional health monitoring often relies on intermittent measurements, making early detection of neurological and physiological changes difficult. NeuroDX AI addresses this challenge by using AI to decode complex multimodal bio signals. Powered by NVIDIA H200 GPUs, TensorRT™, and NVIDIA Triton™ Inference Server, NeuroDX AI accelerates model training and optimizes inference for ultra-low latency and high throughput. Discover how NVIDIA's Inception program enabled NeuroDX AI with specialized training, early technology access, venture capital connections, and global showcase opportunities to turn complex bio signal data into actionable health intelligence.

Real-Time Neurological Insights: Scaling NeuroDX AI with NVIDIA Inception

Praktický pohled na agentickou AI v AI továrně: perzistence, autonomie a sebezokonalování, bezpečnostní zásady, řízení přístupů a výběr modelů pro udržitelnou návratnost v podniku.

AI továrna pro podnik: jak bezpečně nasadit autonomní agenty

Jak bezpečně nasadit autonomní agenty v AI továrně: praktický pohled na agentickou AI, bezpečnost a modely

— Introduction

— Meet our guest: Anne Hecht

— From gen AI to agentic AI: what actually changed

— What broke the dam: enterprises seeing real ROI from their AI projects

— Inside NVIDIA: 4 trillion tokens a day

— Three words to describe AI agents: persistent, autonomous, self-evolving

— The CFO view: tokens, marginal cost, and where AI runs

— The IT & CISO view: manage agents like you manage employees

— "Your agent is only as good as its connections"

— Shadow AI is the new shadow IT

— What's in it for users: privacy, confidentiality, and control

— Easy is hard: designing a seamless agent experience

— The GTC slide: Agent = harness, models, and tools

— Agent skills & why fine-tuned open models cut cost

— Securing agents: the Secure Agent Workspace & OpenShell

— Confidential computing: protecting models & data in use

— Product or reference design? What's what

— Frontier vs. open models: it's an "and," not an "or"

— The Linux parallel: open source as the OS of compute

— Nemotron adoption & post-training

— Jensen Huang on X & the Open Secure AI Alliance

— Wrap-up

In Episode 3 of AI Factory Insider, host Kaushik Shirhatti (VP, AI Factory, NVIDIA) is joined by Anne Hecht (Sr. Director of Product Marketing, NVIDIA) to discuss autonomous agents that continuously learn and compound value, purpose-built AI factory infrastructure, and real-world use cases.

What we cover:
0:00 — Introduction 
1:43 — Meet our guest: Anne Hecht 
3:35 — From gen AI to agentic AI: what actually changed 
5:20 — What broke the dam: enterprises seeing real ROI from their AI projects
6:50 — Inside NVIDIA: 4 trillion tokens a day 
7:17 — Three words to describe AI agents: persistent, autonomous, self-evolving 
11:05 — The CFO view: tokens, marginal cost, and where AI runs 
14:16 — The IT & CISO view: manage agents like you manage employees 
16:28 — "Your agent is only as good as its connections" 
17:59 — Shadow AI is the new shadow IT 
18:51 — What's in it for users: privacy, confidentiality, and control 
19:58 — Easy is hard: designing a seamless agent experience 
21:02 — The GTC slide: Agent = harness, models, and tools 
23:06 — Agent skills & why fine-tuned open models cut cost 
27:24 — Securing agents: the Secure Agent Workspace & OpenShell 
29:23 — Confidential computing: protecting models & data in use 
30:36 — Product or reference design? What's what 
32:52 — Frontier vs. open models: it's an "and," not an "or" 
34:35 — The Linux parallel: open source as the OS of compute 
36:51 — Nemotron adoption & post-training 
37:39 — Jensen Huang on X & the Open Secure AI Alliance 
40:14 — Wrap-up 

Resources:
Secure Agent Workspace Reference Design 👉 https://nvda.ws/3U9E1LE
NVIDIA OpenShell 👉 https://nvda.ws/454Dmh2
NVIDIA AI Factory Validated Design Guide 👉 https://nvda.ws/4wGGHz0
AI Factories for Self-Evolving Agents (Kaushik & Anne's GTC Taipei session) 👉 https://nvda.ws/3TqVyim
Deploying Proprietary Models Securely with NVIDIA Confidential Computing on Self-Hosted VMs 👉 https://nvda.ws/4cgmqI8
Deploying Proprietary Models Securely with NVIDIA Confidential Computing on Self-Hosted Kubernetes 👉 https://nvda.ws/4z0AjnO
Building an Analysis AI Agent for Industrial Alarm Management with NVIDIA Nemotron 👉 https://nvda.ws/4xiIj1N

Securely Deploy Autonomous Agents in Your AI Factory | AI Factory Insider Ep. 3

Zjistěte, jak psát kvalitní prompty, porozumět tomu, jak generativní AI funguje, kdy používat ChatGPT/Claude/Gemini a jak vytvářet opakovatelné AI systémy pro lepší výsledky.

Naučte se promptovat AI: praktický návod a systémy

Jak se naučit používat AI: praktický návod k promptům, systémům a lepším výsledkům

✅ Get Your FREE AI Company Operating System here: https://go.danmartell.com/4uadjA2

🔥 Are you a Business owner? Join my FREE AI CEO Challenge: https://go.danmartell.com/aichallenge2

👥 Are you building an AI software company? Partner with me: https://go.danmartell.com/40anA1K

If AI feels overwhelming, this video will give you the clarity you’ve been missing.

I’ll walk you through the core principles of AI, the essential skills to get good fast, and the steps to future-proof yourself so AI becomes your advantage, not your replacement.

Tools mentioned:

ChatGPT: https://chat.openai.com/
Claude: https://claude.ai/
Gemini: https://gemini.google.com/
DeepSeek: https://www.deepseek.com/
Perplexity: https://www.perplexity.ai/
Grok: https://x.ai/
Slack: https://slack.com/
Notion: https://www.notion.so/

▸▸ Subscribe to The Martell Method Newsletter: https://bit.ly/3XEBXez

▸▸ Get My New Book (Buy Back Your Time): https://bit.ly/3pCTG78

IG: @danmartell

You’re not behind (yet): How to learn AI in 18 minutes

Jak Codex Micro pomáhá vývojářům pracovat s agentním programováním rychleji: hlasové zadávání, fyzické ovládání úloh, přepínání kontextu i plánovací režim. Praktický pohled a ukázka hry One Letter Off.

Codex Micro: programovatelná klávesnice pro hlasové a rychlé ovládání agenta

Codex Micro: programovatelná klávesnice pro rychlejší práci s Codexem

Introducing the Codex Micro

Voice dictation

Skills, Plan mode, and reasoning

Switching between tasks

Notifications and permissions

Queueing follow-up work

Game result and customization

See how the Codex Micro, a customizable keyboard built with Work Louder for Codex, helps you move through a build without leaving your flow. In this demo, GPT-5.6 Sol creates a word game while the Micro handles voice dictation, task switching, reasoning controls, permission requests, and follow-up prompts.

Get yours: https://openai.com/supply/

Chapters:
00:00 Introducing the Codex Micro
00:24 Voice dictation
00:45 Skills, Plan mode, and reasoning
01:01 Switching between tasks
01:16 Notifications and permissions
01:28 Queueing follow-up work
01:49 Game result and customization

Zjistěte, jak převést opakující se pracovní rutiny do naplánovaných úkolů v ChatGPT. Od denního briefingu, přes Chief of Staff až po týmové agenty pro Slack a organizaci kontextu.

Automatizované plánování v ChatGPT: ranní briefing až týmové agenty

Plánování úkolů s ChatGPT: od denního briefingu k týmovým agentům

Scheduled tasks in ChatGPT

Chief of Staff

Creating a daily brief

Automating feedback triage

Reviewing Feedback Triage

Sharing with Workspace Agents

Converting to a cloud agent

See how you can use ChatGPT to automate tasks and create workflows for yourself, then schedule them to run when you need them. Review the results, keep recurring work moving, and share useful workflows with your team.

Learn more: https://learn.chatgpt.com/

Chapters:
00:00 Scheduled tasks in ChatGPT
00:10 Chief of Staff
00:28 Creating a daily brief
00:38 Automating feedback triage
00:57 Reviewing Feedback Triage
01:09 Sharing with Workspace Agents
01:23 Converting to a cloud agent

Schedule tasks with ChatGPT

How Philips scaled AI literacy across 70,000 employees — leadership alignment, ChatGPT Enterprise pilots, company-wide training and an innovation challenge.

Scaling AI Literacy: Philips’ Playbook for 70,000 Employees

How Philips is scaling AI literacy across 70,000 employees

Philips is building AI literacy into a company-wide capability, training executives, running use case challenges, and rolling out ChatGPT Enterprise to 70,000+ employees. Read more: https://openai.com/index/philips/

Nová aplikace Google Health propojuje zdravotní údaje z Fitbitu i Apple HealthKit. Seznamte se s Fitbit Air bez displeje a s Google Health Coach poháněným Gemini, který vysvětluje data srozumitelně.

Google Health s Gemini: AI kouč a propojení zdravotních dat

Jak Google používá AI pro zdraví: nová aplikace Google Health, Fitbit Air a trenér poháněný Gemini

What does building an unmanned helicopter have in common with designing a tiny, screenless fitness tracker? And why did NBA superstar Stephen Curry’s elite athleticism actually break Google’s sleep-tracking algorithm?

On this episode of the Made by Google Podcast, host Rachid Finge sits down with Andy Abramson, Head of Product for Fitbit and Health, to explore the design and engineering behind Google’s newest health tools: the redesigned Google Health app, the screenless Fitbit Air, and the Gemini-powered Google Health Coach.

Andy explains why Google is returning to Fitbit's screenless roots, how the team reacted when Reddit users tried to make their own custom bands, and snapping a few photos can go a long way in keeping your fitness plan on track, thanks to Gemini.

00:38 Complexity in Design: From Helicopters to Fitbit Air
01:31 Redesigning the Google Health App
11:03 Introducing the Gemini-powered Google Health Coach

#MadeByGoogle #TechPodcast 

Subscribe to our Channel: https://www.youtube.com/google
Find us on X: https://twitter.com/google
Watch us on TikTok: https://tiktok.com/@google
Follow us on Instagram: https://www.instagram.com/google
Join us on Facebook: https://www.facebook.com/Google

How We Use AI for Your Health | Made by Google Podcast S9E6

Jak se Codex mění v propojené vývojové prostředí: GPT‑5.6 Sol, režim Ultra, subagentní paralelizaci, Computer Use s appshots, publikování přes Sites, mobilní správa práce a pull request workflow v jedné aplikaci.

Codex v nové éře: GPT‑5.6, Ultra režim, paralelní práce a PR workflow

Codex se právě výrazně zlepšil pro vývojáře: GPT-5.6, Ultra, Sites, mobilní práce a pull requesty v jedné aplikaci

What's new for developers

GPT-5.6 Sol and Ultra mode

Computer Use and appshots

Browser Use and inline editing

Publishing with Sites

Coordinating work across projects

Codex on mobile

Pull request workflows

OpenAI Build Week

Here’s what’s new in Codex for developers. Thanks for helping us make it better!

We packed in a lot: 
* GPT‑5.6
* Ultra
* New ways to manage parallel work
* Big updates to Computer Use 
* Inline code and document edits
* PR reviews
* Sites
* ...and much more.

Make sure you watch until the end: something that isn’t out yet makes a cameo!

Chapters:
00:00 What's new for developers
00:46 GPT-5.6 Sol and Ultra mode
01:28 Computer Use and appshots
02:11 Browser Use and inline editing
03:17 Publishing with Sites
03:51 Coordinating work across projects
04:57 Codex on mobile
05:36 Pull request workflows
06:35 OpenAI Build Week

Archive