The Agent Arms Race Heats Up

Edition 141 | December 4, 2025

Bland AI’s robot is out here living a full Hollywood romance arc (golden-hour hug, dramatic score, slow-motion everything), and the average AI agent is stuck in call-center purgatory, whispering sweet nothings like “How may I redirect your call?”

Welcome back to Building AI Agents, your biweekly guide to everything new in the field of agentic AI!

In today’s issue…

• OpenAI triggers “code red” as Gemini 3 closes the gap

• AWS unveils “billions of agents” strategy

• Amazon launches frontier agents including Kiro

• MIT’s Iceberg Index shows 151M U.S. workers exposed

• How Vercel replaced a $1M SDR team with a $1K agent

…and more

🔥 INCASE YOU MISSED IT

Readers’ favorite items from the past week

1. 5 day agent building course by Google

2. IRS deploys Salesforce AI agents

3. Practical lessons when building and deploying agents

4. 5,000 MCP servers for every use-case

   disclaimer: MCP servers can be security risks—make sure you know what you’re connecting to!

5. Gemini 3 Prompting Guide

📌 THE BRIEFING

🚨 OpenAI's "Code Red" Signals a Shifting AI Landscape

Sam Altman has declared a "code red" at OpenAI, redirecting resources to urgently improve ChatGPT as Google and Anthropic close the gap. In an internal memo (obtained by the WSJ), Altman told employees that OpenAI is delaying work on AI agents, advertising, and other initiatives to focus on ChatGPT's speed, reliability, and personalization.

The urgency follows Google's Gemini 3 release, which surpassed ChatGPT on benchmarks and drew public endorsements from leaders like Salesforce CEO Marc Benioff — a stark reversal from three years ago when Google sounded its own alarm over ChatGPT's launch. For agent builders, the foundation model wars are heating up again, and the resulting competition should accelerate capabilities across the board heading into 2026

☁️ AWS Bets Everything on "Billions of Agents"

At re:Invent 2025, AWS CEO Matt Garman declared that autonomous agents will reshape enterprise computing as profoundly as the internet and cloud. Garman told attendees that "80 to 90% of enterprise AI value will come from agents," envisioning billions operating across every company and industry.

AWS backed up the vision with concrete releases: "frontier agents" including Kiro (an autonomous coder that works for days without intervention), Bedrock AgentCore for deploying agents at scale, and a Visa partnership enabling agentic commerce. The message to builders is clear: the infrastructure layer is now being optimized specifically for agentic workloads.

🤝 WITH SYNTHFLOW

The New Framework for Enterprise Voice AI

Enterprise teams are automating more calls than ever — but without a consistent framework, deployments become unpredictable, costly, and slow to scale.

The BELL Framework introduces a structured way to design, test, launch, and improve Voice AI agents with reliability.

Get the guide enterprises are now using to de-risk voice automation and accelerate deployment.

Download The Free Guide

 🤖 AGENT OF THE WEEK

👋 Welcome back to Agent of the Week!

I run an ad agency for local businesses, and one of the biggest challenges isn’t launching campaigns — it’s the resources and time spent auditing them.

My team has to pull reports, analyze wasted spend, hunt for bad search terms, check radius performance, and review which headlines are dying.

It’s repetitive.
It’s time-consuming.
And it’s exactly the kind of work AI should be doing for us, according to my B.O.T.S Framework.

So I started experimenting with Google Gemini 3 Pro to see if it was good at analyzing Google Ads. I mean.. it is a google product.

I crafted a prompt, dropped in the four raw Google Ads reports, and let Gemini run a full audit — the kind we normally spend hours doing manually.

And honestly? it was actually good.
This prototype is now on my list to become a fully autonomous internal agent for our team.

This is the kind of concept-to-reality work I love. You run some experiments semi-manually to test the concept, and if it proves itself useful, you turn it into a real agent.

So here’s how it works:

🧠 How It Works

1. Copy & Paste this → Prompt & Drag in the Google Ad Files to Gemini 3 Pro
4 Files: Search Terms, Locations, Schedule, and Assets as CSVs. Gemini ingests, cleans, and links all the data automatically

2. The Bleed Audit → Wasted Spend Finder
It scans every search term, flags spend with 0 conversions, clusters the junk queries, and generates a 10-word “Negative Keyword Kill List”

3. The Radius Audit → CPA by Distance
Gemini analyzes performance at 0–5 mi, 5–10 mi, and 10+ mi, and identifies the exact distance where your CPA becomes unprofitable

4. The Vampire Audit → Time-of-Day Leak Detection
It compares business hours vs off-hours, finds nighttime/weekend waste, and gives you a clean “Ad Schedule to Turn Off”

5. The Creative Audit → Winners vs. Losers
Every headline, description, and sitelink gets graded. Gemini surfaces the winning creative to keep — and the losers to delete

6. Final Synthesis → Strategy Report
All four audits fuse into a single “Start / Stop / Continue” plan with total wasted spend, optimization steps, and the highest leverage actions to get more ROI

Now I built this for myself, but…

You can modify the prompt if you run a different type of business or have different items you want analyzed for your google ads.

The base layout is all there, and if you have any question please reply to this email and just ask me!

For more tip & trick and learnings on how to build your own no-code agents, join the Building AI Agents community — 7 day free trial!

Till next week,
✌️ AP