• Everyday AI
  • Posts
  • Inside Multi-Agent AI: Rethinking Enterprise Decisions

Inside Multi-Agent AI: Rethinking Enterprise Decisions

GPT-4.5 released, Amazon unveils quantum chip, NVIDIA revenue soars, Microsoft calls for removal of AI chip restrictions and more!

Author

Everyday AI
February 27, 2025

Sup y’all 👋 

More model drops the past 7 days than a Paris Fashion show on a tight wire. Grok, Sonnet and now we’ve got GPT-4.5 from OpenAI. (Well, for Pro users only for now.)

Make sure to join us LIVE at 7:30 AM CST. If you can’t make it, go leave us a question to tackle live on the stream/podcast.

✌️

Outsmart The Future

Today in Everyday AI
7 minute read

🎙 Daily Podcast Episode: Once AI agents are everywhere, how will enterprises navigate multi-agent AI? We dive in and find out. Give it a listen.

🕵️‍♂️ Fresh Finds: Microsoft Copilot now available on MacOS, Ideogram releases new AI model and NVIDIA CEO speaks on AI computing. Read on for Fresh Finds.

🗞 Byte Sized Daily AI News: OpenAI launches GPT-4.5, Amazon unveils new quantum chip and Microsoft calls for the removal of AI chip restrictions. For that and more, read on for Byte Sized News.

🚀 AI In 5: We dive into Sora for a quick review treatment. Is Sora as good as when it was first announced? See it here

🧠 Learn & Leveraging AI: Here’s what you need to know about multi-agent AI to prepare and grow your company or career. Keep reading for that!

↩️ Don’t miss out: Did you miss our last newsletter? We talked about Amazon unveiling an AI-powered Alexa+, Microsoft’s next-gen Phi AI models and OpenAI launching GPT-4.5 soon. Check it here!

 Inside Multi-Agent AI: Rethinking Enterprise Decisions 💡

What happens when.... AI agents are everywhere?

To learn, we tapped into the insights from one of the leading voices in AI, Babak Hodjat, whose resume includes helping create the tech behind the original AI agents like Siri.

So, how do enterprises prepare for a multi-agent environment? We dive in to find out.

Join the conversation and ask Jordan questions on AI agents here.

Also on the pod today:

• Implementing Multi-Agent Systems 👥
• Hallucinations and Errors in AI Systems ❌
• Usage and Organization within Multi-Agents 🗂

It’ll be worth your 37 minutes:

Listen on our site:

Click to listen

Subscribe and listen on your favorite podcast platform

Listen on:

Tomorrow’s Podcast:

Here’s our favorite AI finds from across the web:

New AI Tool Spotlight – Polymet is an AI product designer, Magic Inspector is an AI web test automation platform and Nuvio provides AI-powered financial management.

Microsoft – Copilot is now available for MacOS.

Redditors have found that Copilot can help pirate Windows 11.

Money in AI - IBM has acquired HashiCorp for $6.4 billion.

NVIDIA – NVIDIA’s CEO says AI has 100 times more computation than when ChatGPT was released.

AI Models - Inception Labs has released Mercury, a commercial grade diffusion LLM.

AI Images – Ideogram has released Ideogram 2a, its newest text-to-image model.

1. OpenAI Unveils GPT-4.5 🚀

OpenAI has officially launched GPT-4.5, touting it as its most knowledgeable model yet, although not a frontier model. With significant improvements in writing capabilities and world knowledge, this version aims to offer a more natural interaction experience, making it ideal for tasks like programming and problem-solving.

However, it’s worth noting that while GPT-4.5 reduces hallucinations compared to its predecessor, GPT-4o, it still trails behind some other models in performance metrics. Following its debut for ChatGPT Pro users, wider access will roll out in the coming weeks, setting the stage for future advancements with GPT-5 expected later this year.

2. Amazon Unveils Quantum Chip Ocelot 🧠

Amazon Web Services has launched its first quantum computing chip, Ocelot. Developed in collaboration with Caltech, this chip promises to significantly cut costs related to error-correction by up to 90%, potentially accelerating advancements in quantum computing for real-world applications.

As quantum technology continues to evolve, the implications for industries such as drug discovery and AI are substantial, hinting at transformations that could redefine how businesses operate.

3. Nvidia Revenue Soars 78% Amid AI Surge 📈

Nvidia's latest earnings report showcases a staggering 78% revenue increase, reaching $39.3 billion, fueled by insatiable demand for AI GPUs, particularly from cloud service giants like AWS and Google Cloud. With the successful launch of the Blackwell GPU generating $11 billion in its first quarter and the anticipated release of the more powerful Blackwell Ultra later this year, Nvidia is poised to maintain its leadership in the AI hardware market.

However, geopolitical challenges loom, as U.S. chip export restrictions have halved Nvidia's revenue from China, raising questions about future growth.

4. Microsoft Calls for Easing AI Chip Export Restrictions 👀

Microsoft is urging President Trump's team to lift export restrictions on artificial intelligence chips that could disadvantage U.S. allies like India, Switzerland, and Israel. The tech giant argues that these measures threaten American tech companies' ability to expand AI infrastructure globally while inadvertently giving China an edge in the AI race.

With market leader NVIDIA facing potential setbacks from these curbs, the pressure is on to reconsider the Biden administration's strict rules introduced earlier this year.

5. Tencent Takes a Shot at DeepSeek with Turbo S 🥊

Tencent has launched the Hunyuan Turbo S, claiming it can answer queries in under a second—outpacing rivals like DeepSeek's R1. This development comes on the heels of DeepSeek's remarkable rise, which has prompted major players like Tencent and Alibaba to hastily upgrade their offerings to stay relevant.

With Turbo S reportedly matching DeepSeek-V3 in knowledge and reasoning tasks while also slashing usage costs, it seems the pressure is mounting on Chinese tech giants to innovate rapidly.

6. Microsoft Targets Cybercriminals in AI Misuse Case 🧑‍⚖️

Microsoft has ramped up its legal efforts by naming four developers linked to a global cybercrime network accused of bypassing safeguards on generative AI to create illicit deepfakes of celebrities.

The tech giant asserts these individuals exploited compromised accounts to not only evade security measures but also sell access to harmful content generation instructions. This action marks a significant step in Microsoft's ongoing battle against AI misuse, aiming to deter others from similar misconduct.

OpenAI Sora Quick Review

Remember that Sora thing?

Which in AI years feels like it was like 13 years ago?

Well it’s now been available to the public and the hype kinda died down.

Was it worth the wait?

🦾How You Can Leverage:

10-15 minutes.

That's how long it now takes Cognizant to create an AI proof-of-concept that used to require 10-12 weeks.

Not a typo. Not hype. A 99.7% timeline reduction happening TODAY in a 360,000-employee company through "agentification."

We sat down with Babak Hodjat, CTO of AI at Cognizant and co-inventor of Siri's natural language tech, who's been crafting multi-agent systems since the 90s. 

While everyone else frets over hallucinations, he's building AI networks that are transforming enterprise decision-making at warp speed.

That shiny new AI strategy you just finalized? 

Already collecting dust, y’all. 

Here’s what you need to know from today’s show. 

1 – The 672X Accelerator: Agentification 🤖

Multi-agent systems don't just improve single AIs—they reshape organizational structure from the ground up.

Babak's team at Cognizant shrunk implementation time from 10-12 weeks to 10-15 minutes by replacing monolithic modules with interconnected agents. 

That's 672 times faster than your current approach.

(Like, we’re not that good at math, but that math maths, right?) 

When Babak typed "my son just turned 26," the system instantly recognized cross-departmental implications—adjusting payroll, modifying benefits, even offering celebration time off. No forms. No tickets. No bureaucracy.

The secret isn't better AI, Babak said. The key is breaking AI into specialized pieces with clear responsibilities and seamless communication.

Try This: 

Identify one cross-departmental process that creates employee frustration. Map every decision point, ownership, data requirements, and system dependencies.

 Redesign assuming each decision had its own specialized agent with clear autonomy boundaries. Create this "agent blueprint" and implement a proof-of-concept connecting just two decision points. 

Track not just time saved but reduction in employee headaches and context-switching.

2 – The Compounding Risk: When 99% Accuracy becomes a threat 🎯

Everyone obsesses over single-LLM hallucinations. 

Nobody's talking about the far more dangerous compounding errors across connected agents.

An agent that's 99% accurate sounds amazing. But when that output becomes another agent's input, then another's? Errors compound exponentially.

The wild part? Babak suggests MORE AIs might fix AI inconsistency.

Multiple agents cross-checking can catch errors before they cascade. 

One agent's hallucination becomes obvious when three others disagree.

But this contradicts the efficiency narrative driving AI adoption. It requires verification layers that might slow systems but make them bulletproof.

Most AI governance frameworks weren't built for this new reality.

Try This: 

Run an "error propagation test" immediately. Design five subtly flawed inputs and track how far incorrect information travels before someone catches it.

Establish "confidence thresholds" with automatic human escalation when scores drop below critical levels. 

Run this quarterly as you add more automation. Share results transparently to build organizational awareness of both AI and human failure modes.

3 – One Employee + Five Agents > Five Employees 🦾

Babak revealed how he automated his email workflow by creating an entire virtual team mirroring his organization.

His AI team didn't just sort his inbox—they analyzed content, made judgment calls, routed messages to the right places, and even drafted responses that sounded exactly like him.

This isn't one assistant. It's a personalized organization of specialized AIs working together to multiply your capabilities.

Success won't come from working faster. It'll come from becoming an AI conductor—orchestrating specialized agents while you focus on truly human-centric value.

The future knowledge worker doesn't do the work—they direct an AI team that does. Their competitive edge shifts from personal execution to effective delegation of intent.

The skills gap between those who can and can't direct AI teams will make today's digital divide look like a hairline crack.

Try This: Build your personal agent network around one recurring workflow that drains your energy. 

For email, create three distinct agent personas: a Triage Specialist categorizing by urgency, a Research Analyst gathering context, and a Communications 

Drafter preparing aligned responses. Give each specific guidelines about what they handle independently versus when to involve you. Use existing tools but customize with precise role descriptions. 

Start with low-stakes tasks and document which instructions yield the best results, creating your personal playbook for effective agent orchestration.

The multi-agent revolution isn't coming—it's here while everyone's still debating prompt techniques. 

You ready? 

⌚

Numbers to watch

$30 Million

Unique, an AI platform for finance has raised $30 million.

Now This …

Let us know your thoughts!

Vote to see live results

Do you attend our livestreams?

Every weekday, we bring you fresh AI insights, exclusive interviews, and breaking news with our Everyday AI livestream.

Login or Subscribe to participate in polls.

Reply

or to participate.