PRODUCTS

Amazon makes AI agents easier to use and trust

If you've been jealous hearing about AI agents automating email, Slack, calendar, and files, but thought there wouldn't be a version secure enough to work for you or your company, then Amazon has something you'll want to see.

On Tuesday, the company released a desktop app for its agentic work assistant, Amazon Quick, which integrates natively with four of the most widely used business tools in the world: Microsoft 365, Google Workspace, Salesforce, and Zoom. Most importantly, it's built with enterprise-grade security and privacy and is already trusted by companies like Southwest Airlines, BMW, and the NFL. 

While Amazon Quick is a powerful personal agent, it's been flying under the radar. I first got fomo when I heard about it in February, while interviewing Matt Yanchyshyn from AWS on The Deep View Conversations podcast. Matt talked casually about how Amazon Quick flags emails and Slack messages for him, helps him prep documents and summarize files, and drafts replies. 

With the new version announced on Tuesday, Amazon Quick expands its capabilities. Here are some of the highlights:

• Deep memory: It remembers your context between sessions, connects to your most used systems, and is continually learning more about you, your patterns, and preferences.

• Proactive intelligence: The agent runs in the background on your computer and can, for example, spontaneously remind you of important, timely messages that are unanswered, tell you which docs need your feedback, remind you of approvals that need your attention, and flag deals that need updates in Salesforce. 

• Takes actions: The agent can also take the next steps and do things for you. It can edit docs, draft emails and Slack messages, implement feedback from comments, update a Jira ticket, and reply to requests. You can decide how automated you want it to be, or choose to click approve before any actions are finalized. 

• Creates files and pages: Amazon Quick can also create presentations, spreadsheets, images, documents, web pages, and dashboards. Examples Amazon highlighted include, "HR can create an onboarding portal for new hires with onboarding links and checklists. Finance can launch a resource/budget calculator. Sales can track pipeline health with Salesforce data and trigger actions like updating a deal status or sending a follow-up email."

You can now download Amazon Quick and try the free version. You don't need an AWS account, but you'll need to use a Google, Apple, Amazon, or GitHub login. 

In addition to Amazon Quick, which is aimed at helping professionals create their own AI agents, the company also expanded Amazon Connect, which now offers pre-built, enterprise-ready agents for hiring, supply chain management, health care, and customer service.

In the past couple of months, I can't tell you how many times I've joked with my colleagues, Nat Rubio-Licht and Sabrina Ortiz, "I really need an AI agent to do this for me" for various tasks. But as much as we test different AI tools and push ourselves to work on the cutting edge of agents, there's still something scary about turning over your most private and confidential work messages and files to a non-deterministic system. It's tough for many companies and professionals to trust data as important as that to a startup or an open-source project because they want someone to call (or blame) if things go sideways. Since AWS has earned a lot of trust and goodwill as a cloud partner for organizations of different shapes and sizes, there are likely a lot more folks willing to give Amazon Quick a try for personal AI agents in business. I certainly intend to see if it's as simple and intuitive as it sounds. That would be a huge win. 

Disclosure: Jason Hiner's travel to What's Next With AWS was paid by Amazon. The Deep View's coverage is editorially independent from the companies we cover.

IN PARTNERSHIP WITH LAMBDA

Cut your AI training costs by 25% or more

Most large-scale AI training runs use less than half the computing power they're paying for. Lambda's team found the root causes and built a reproducible framework that boosted efficiency by over 25%, without changing the model itself.

Lambda’s whitepaper shows you how to address:

• Memory inefficiencies silently inflating your costs

• Training configurations that aren't making full use of your hardware

• Bottlenecks that slow down GPU communication

Get the guide.

ENTERPRISE

Perplexity shifts its desktop agent to enterprise

Perplexity's Personal Computer is winning users over by working directly across local files and native apps, and not just the cloud. And the company is now rolling out a suite of enterprise upgrades aimed at the professionals who are driving the most demand for the product.

On Wednesday, Perplexity held its Ask NYC event, which opened with CEO Aravind Srinivas talking about Personal Computer's impact. Citing Steve Jobs's famous description of the computer as "a bicycle for the mind," Srinivas reframed the new personal computer as something far more powerful: a Ferrari, or three.

Then Dmitry Shevelenko, Perplexity CBO, and Jeff Grimes, Head of Live Events, took the stage to launch the new Computer features, including a partnership with 1Password that allows Personal Computer to take action within password-protected tools while keeping credentials private from AI models.

Another large focus was broadening access to Computer. Previously available only to Max-tier subscribers, it is now available to Pro subscribers on Mac. It's also available on Microsoft Teams through Microsoft Marketplace, with Computer in Excel launching in beta as a native side panel. 

Computer will now also offer:

• Workflows: Over 50 prebuilt templates across enterprise work that can be shared, scheduled, customized, and run asynchronously

• Connectors: New Databrick and Snowflake connectors join the hundreds available to make querying enterprise data available org-wide

Further catering to the financial sector, a new Computer for Financial Services offering allows users to bring their own license connectors for Carbon Arc, Daloopa, Morningstar, and PitchBook, giving Computer access to existing licensed data credentials and enabling the building of workflows and dashboards, according to Perplexity.

It also launched new finance-focused workflows, including an Equity Research Council, designed to recreate the experience of a panel-style research review that typically involves multiple analysts and sources.

Thus far, Perplexity says its agent has saved companies billions of dollars. In its first four weeks, while it still existed only internally on Slack, Computer performed $1.6 million worth of work, according to Shevelenko. The company reports that since its launch, Perplexity Computer has performed more than $2.8B in labor-equivalent work.

Perplexity has been especially nimble at adapting to the AI landscape. When it first launched, it was mainly a capable AI search engine, with a primary focus on consumers. Over time, it not only expanded its AI offerings far beyond search but also captured serious attention from enterprises. Its pivot has been faster than the one OpenAI is now trying to make, likely because it has been agile in meeting AI demand where it is. Personal Computer has been a perfectly timed answer to the OpenClaw viral moment, while offering an easier on-ramp and rapidly scaling up its enterprise capabilities.

IN PARTNERSHIP WITH ORACLE NETSUITE

Guide: Financial storytelling meets AI

Your job isn’t just presenting numbers—it’s getting leaders to interpret what matters and decide what to do next. Don’t let strong analysis end with a polite nod.

Download Financial Storytelling in the Age of AI to apply the “Made to Stick” SUCCES(S) framework—plus AI-assisted drafting—to make messages clearer, more credible, and easier to remember.

Get Your Guide

BIG TECH

Nvidia brings agents closer to real work

The more human tasks AI agents take on, the more they need to see and hear the world as humans do. Nvidia's new model is designed to bridge that gap.

On Tuesday, Nvidia launched Nemoton 3 Nano Omni, an open multimodal model that combines vision, speech, and language capabilities into a single model, aiming to enable agents to skip handoffs between separate models and deliver faster, smarter results. 

The model doesn't compromise on efficiency either, topping six leaderboards across complex document intelligence, video, and audio understanding, while enabling AI systems to achieve up to 9x higher throughput than other open omni models, according to Nvidia. This performance is enabled by: 

• Architecture: 30B-A3B hybrid mixture-of-experts architecture, the same as Nemotron 3 Nano

• Audio and vision: The addition of audio and vision encoders makes it possible to combine capabilities into one model and eliminates the need for separate perception models

• Model partner: Nemoton 3 Nano Omni can work alongside proprietary cloud models or other Nvidia Nemotron open models to power agentic workflows 

Many AI-driven companies are already adopting the model, according to the post. Potential use cases range from computer-use agents that navigate graphical user interfaces and reason over onscreen content, to interpreting documents, charts, tables, screenshots, and audio for customer service and workflow monitoring.

The model is available now through Hugging Face, OpenRouter and build.nvidia.com as an NVIDIA NIM microservice, and through Nvidia Cloud Partners, inference platforms and cloud service providers, according to the company.

As AI agents continue to dominate the AI hype cycle, the technology must evolve to meet growing demand and the increasingly complex needs of these models. Nvidia is well-positioned to contribute here, not only because it has the research talent and engineering prowess to drive meaningful advances, but also because its primary business is in chips. As TDV's Jason Hiner noted previously, that independence gives it the freedom to build models that serve a broader need, rather than being beholden to any particular customer.

LINKS

• Musk testified he is suing OpenAI for the for-profit business pivot

• OpenAI’s Codex tool has a line that prevents mention of globlins

• AI by Mayo Clinic can identify pancreatic cancer years before it shows up

• Rogo, AI-startup founded in 2021 by bankers, notched a $2 billion valuation

• MIT Researchers develop method for privacy-preserving AI training 

• MIT-IBM Computing Research Lab will research AI and quantum

• QA Tech: Delegate testing to AI. Validate releases with dynamic regression and exploratory tests run by agents that act like your customers. Try QA.tech now. (sponsored)

• Copilot in Outlook: Microsoft rolled out new agentic experiences for email and calendar 

• Motubrain: New World Action Model from ShengShu Technology

• ElevenMusic: ElevenLabs launched ElevenMusic, a platform for creating music with AI 

• CursorSDK: Users can build agents with same runtime, models, etc. that power Cursor.

• Manus Slides: Now gives users GPT Image 2 option to create and edit slides 

• Gemini: Users can now ask Gemini to create Docs, Sheets, Slides, PDFs, in chat

• Microsoft: Principal Software Engineer - Azure AI Translation & Language Team 

• HPE: Private Cloud AI Customer Engineer

• Meta: Research Engineer, Monetization AI

• Snowflake: Staff Research Scientist, AI Agents & LLMs

GAMES

A QUICK POLL BEFORE YOU GO