AI news don’t stop, and neither do we. 

Let's get into this week's AI feast, shorties.

Here’s what ya need to know. 👇

Microsoft just fired a cruise missile directly at Salesforce's headquarters with their new AI sales agents.

Sales Agent and Sales Chat can identify leads, schedule meetings, follow up with customers, and analyze data without human handholding.

The SAVAGE part? They integrate with BOTH Microsoft Dynamics 365 AND Salesforce, letting sales teams bypass opening their CRM entirely.

These AI agents enter public preview in May 2025 via Microsoft 365 Copilot, transforming from mere prompts to fully autonomous task-crushers.

Microsoft's also launching an AI Accelerator program to fast-track businesses onto their platform.

They're low-key trash-talking "legacy CRM vendors" (Salesforce, we see you) who picked this fight with Microsoft. (We think it was just for free marketing/visibility)

What it means

Microsoft's diabolical genius move lets their AI infiltrate both platforms, which could turn Salesforce into a glorified database while Microsoft's intelligence handles the ACTUAL relationship management.

Sales teams will abandon traditional CRM interfaces faster than crypto prices after an Elon tweet – nobody enjoys data entry, and now they won't have to.

The real question will be which AI sales agent wins: this one, or Salesforce’s Agentforce. Time will tell shorties.

Ex-Google DeepMind researchers just secured a MASSIVE $130M to build what comes after AGI.

The cash avalanche came in two waves: $25M seed led by Sequoia Capital and CRV, then an immediate $105M Series A co-led by CRV and Lightspeed.

Their investor roster reads like an AI dream team: NVIDIA's VC arm, LinkedIn's Reid Hoffman, and Scale AI's Alexander Wang are ALL in.

First target: autonomous coding tools that handle complex programming without human intervention.

Their sneaky strategy: mastering coding creates breakthroughs in reasoning and self-improvement that naturally extend to general intelligence.

In other words, a coding agent that teaches itself becomes the foundation for systems smarter than humans.

YIKES.

What it means

The superintelligence could be over before it even starts between Reflection AI and Safe Superintelligence Inc.

These DeepMind veterans aren't raising $130M for cute chatbots – they've seen what's coming behind closed doors and they're SPRINTING toward superintelligence while regulators tie their shoes.

Chinese startup Monica just unleashed an autonomous beast called Manus that aims to makes ChatGPT look like a pocket calculator.

This digital demon handles EVERYTHING without human handholding – browsing, coding, image generation, and even completing freelance gigs on Upwork and Fiverr completely autonomously.

Manus combines Claude 3.5's transformer architecture with Alibaba's QWQ reasoning capabilities in an unholy alliance of AI power.

It's crushing both ChatGPT and Google Gemini on benchmarks, according to the team.

The SCARIEST part? It fuses OpenAI's Deep Research-esque capabilities with Operator-like autonomy in one system. And a bunch more.

Currently invite-only, but they're planning to open-source the models in 2025, unleashing chaos on the AI landscape.

What it means

When a Chinese startup combines Claude with Alibaba's reasoning and could outperform America's finest outta nowhere, we're witnessing the digital version of Sputnik.

…..

If it works.

This could be a small wave. Or it could be a moment that reshapes global tech power dynamics FOREVER.

(We’re guessing it’ll be somewhere in between.)

Meta's head of business AI Clara Shih just dropped a big number: they're targeting HUNDREDS OF MILLIONS of businesses with AI agents.

Their tech already reaches 700 MILLION individuals through Facebook, WhatsApp, and Instagram – that's more than twice the entire US population.

While OpenAI chases Fortune 500 paychecks, Meta's courting the little guys – small businesses with zero in-house AI talent.

Meta’s upcoming AI minions will handle customer questions, remember preferences, and automate boring tasks 24/7 inside apps businesses ALREADY use.

LlamaCon drops in April with whispers of Llama 4 and new autonomous agents coming.

What it means

Meta's about to DEVOUR the small business AI market while OpenAI sips champagne with enterprise execs.

Their strategy is LETHAL – embedding affordable AI into platforms already used by billions creates an ecosystem small businesses will flock to like teenagers to TikTok.

Anthropic just told the White House to WAKE UP because superhuman AI is coming NEXT YEAR.

In a 10-page policy smackdown, they asked for classified communication channels between AI labs and the government – like, YESTERDAY.

They want top researchers getting expedited security clearances so government officials aren't blindsided by what's coming.

CEO Dario Amodei didn't mince words, comparing the situation to ARMS CONTROL coordination. Let that sink in.

The bombshell: AI matching Nobel-level human intellect drops in 2026. Not 2030. Not 2040. NEXT YEAR.

This creates massive tension between national security needs and public transparency about what's really happening in AI labs.

What it means

Anthropic is SCREAMING that extinction-level AI risks are imminent while the government has its fingers in its ears.

When a leading AI lab publicly demands classified government oversight channels for technology arriving next year, they're telling us our current safety measures are LAUGHABLY inadequate.

The Trump administration's regulatory vacation means we're racing toward superhuman AI with ZERO guardrails.

Prolly not good.

Google just detonated traditional search with AI Mode in their Google One AI Premium service.

Built on Gemini 2.0, it doesn't just give you links – it delivers COMPLETE answers to your questions.

Kinda like a cross between its traditional AI Search, a lil AI chatbot, and Perplexity.

The game-changer: users can have actual conversations with search, asking follow-ups and diving deeper without starting over.

It queries MULTIPLE websites simultaneously and explores related topics automatically – like having a research team at your fingertips.

Google actually programmed HONESTY – it refuses to generate answers when it's uncertain. Revolutionary concept in AI!

But the current version shows fewer links than promised in demos, sending PANIC waves through the publishing world. Rut-roh?

What it means

Google finally admitted what everyone knows: people want ANSWERS, not ten pages of keyword-stuffed garbage written by underpaid content farms.

(Sorry not sorry.)

Alibaba just dropped a TINY model that's making its Western LLM countparts (and DeepSeek!) look like overpriced gas-guzzlers.

QWQ-32B ("Qwen with Questions") is fully open-sourced under Apache 2.0 – FREE for anyone to use or modify.

The MIND-BLOWING part? With reportedly only 32 billion parameters, it's matching DeepSeek's R-1 with 67 BILLION parameters.

Their secret weapon: advanced reasoning that generates intermediate questions during inference – it literally asks itself questions to improve its thinking.

It's DOMINATING math and coding benchmarks, tying or beating OpenAI and DeepSeek in FIVE major reasoning tests.

Within DAYS of launch, it claimed the #1 spot on open-source leaderboards, sending Alibaba's stock to the MOON.

(Not actually to the moon, but…. yeah, their stock went up. lolz)

What it means

China just SCHOOLED the West on AI efficiency while Silicon Valley was busy building models that need their own power plants.

If it all checks out, Alibaba proving a 32B model can outperform systems 20× larger is like showing up to an arms race with a laser gun while everyone else is still perfecting bigger cannons.

ChatGPT's Mac app just got a STEALTH upgrade that's about to make coders lose their minds.

It now works DIRECTLY with IDEs like Xcode and VS Code – not just suggesting code but writing and modifying it automatically inside your actual development environment.

The killer feature that NO ONE is talking about: "auto apply" mode that implements changes without the copy-paste dance.

This is available to all paid users – Plus, Pro Teams, and Enterprise – as part of the "Work with Apps" capability.

This follows Claude Code's release but with a KEY difference. ChatGPT's version needs no terminal know-how, just a paid account.

The days of switching between your IDE and AI assistant are OVER – ChatGPT lives where you code now.

What it means

Traditional coding interfaces just got VAPORIZED overnight.

Pew pew.

This update transforms ChatGPT from "helpful suggestion box" to "pair programmer on steroids," eliminating ALL the friction that kept AI assistants from taking over development workflows.

OpenAI's about to rent out AI EMPLOYEES at prices that would make a Ferrari dealer blush.

The Information reports THREE tiers coming: $2,000/month for business tasks, $10,000/month for software development, and an absolutely BONKERS $20,000/month for "PhD-level" research agents.

These aren't cute chatbots – they're fully autonomous digital workers that operate independently on your behalf.

SoftBank has reportedly committed THREE BILLION DOLLARS to this venture for 2025. That's "billion" with a "B".

OpenAI expects these agent services to eventually make up 25% of their total revenue – a MASSIVE business pivot.

Sam Altman wasn't kidding when he predicted AI agents would "join the workforce" this year.

He just meant at executive salary levels.

What it means

OpenAI just dropped the pretense and admitted they're SELLING INTELLIGENCE to the highest bidders.

(Anyone wanna chip in. Who’s got like $19,800 a month to spare?)

Their pivot from democratizing AI to renting digital employees at luxury car prices exposes the REAL business model – creating a two-tier AI economy where only enterprises get the good stuff.

The truly WILD part?

The math STILL maths.

A 24/7 PhD-level digital researcher at $20K monthly is CHEAPER than hiring three human researchers at $200K+ annually each. Corporate America will line up with credit cards out.

Microsoft is building AI escape hatches despite reportedly owning nearly HALF of OpenAI. The DRAMA!

Although Copilot has primarily relied on OpenAI’s GPT models since launch, that could soon be changing.

According to reports, their internal research team has created "MAI" reasoning models that nearly match OpenAI and Claude on benchmarks – right under Sam Altman's nose.

The JUICIEST part: Microsoft's already testing Copilot with Elon's Grok, Meta's Llama, and even Chinese model DeepSeek.

This isn't just casual dating – it's about MONEY.

Running GPT-4.5 in future Copilot models would burn cash faster than a Silicon Valley startup, and Microsoft would foot that bill for 365 Copilot.

With their own top-tier models, Microsoft could strong-arm better terms from OpenAI or use a mixture of models instead of JUST OpenAI’s.

No official breakup announcement, but this relationship status just changed to "it's complicated."

What it means

Microsoft just put OpenAI on NOTICE that they're not the only AI in the sea.

Building competing models while openly testing EVERY alternative in town is the corporate equivalent of updating your dating profile while still in a relationship – a CLEAR signal that Microsoft refuses to be held hostage by OpenAI's pricing or roadmap.