• Everyday AI
  • Posts
  • Ep 762: Agentic Context Carry: 3 Steps to Improve Cowork and scheduled AI Workflows (Start Here Series Vol 22)

Ep 762: Agentic Context Carry: 3 Steps to Improve Cowork and scheduled AI Workflows (Start Here Series Vol 22)

ChatGPT’s new GPT-5.5 spud model drops, Microsoft made Copilot agentic across Office apps, and White House accuses China of stealing U.S. AI.

Author

Everyday AI
April 23, 2026

 

Sup y’all 👋

OpenAI JUST released GPT-5.5 and it looks like a BANGER!

(They have the top 3 spots on Artificial Analysis, which has never been done before to my knowledge.)

The model looks MUCH more agentic by default, as demos showed simple real-world work examples like, “Turn this Q3 financial forecast into a presentation” and then 5.5 uses the new image model to create a BANGER deck.

(Glad they stopped doing ‘buy a concert ticket’ use-cases for this launch. lolz)

What’s your take?

GPT-5.5: What's your take?

🗳️ Vote to see LIVE results 🗳️

Login or Subscribe to participate in polls.

✌️

Jordan

Outsmart The Future

Today in Everyday AI
8 minute read

🎙 Daily Podcast Episode: OpenAI and Google just released agents that connect to your apps, carry context between them, and run recurring work automatically. Give today’s show a watch/read/listen to learn more.

🕵️‍♂️ Fresh Finds: New GPT-5.5 takes top 3 spots on Artificial Analysis rankings, OpenAI launched a new Privacy Filter model, Google says an engineer stole AI secrets and sent them to China, and more. Read on for Fresh Finds.

🗞 Byte Sized Daily AI News: ChatGPT’s new GPT-5.5 spud model drops, Microsoft made Copilot agentic across Office apps, and White House accuses China of stealing U.S. AI. Read on for Byte Sized News.

💪 Leverage AI: AI agents can now wake up on a schedule, pull from your email, calendar, Slack, and files, and hand you finished work before your day starts. Keep reading for that!

↩️ Don’t miss out: Miss our last newsletter? We covered: OpenAI just launched ChatGPT Images 2.0, Anthropic’s Claude Mythos was briefly accessed in a leak, Google turned Gemini into a full enterprise agent platform, and more. Check it here!

Ep 762: Agentic Context Carry: 3 Steps to Improve Cowork and scheduled AI Workflows (Start Here Series Vol 22)


Info hunting and juggling sound familiar?

It’s the downfall of almost any business leader. Where is that email from Emily? Why can’t I find last quarter’s budget in Drive? Oh, and Keenen needs an answer back on that research project. Oh shoot, I swear Caleb confirmed the expenses in one of these Slack channels.

You’re off an information rabbit hole and by the time you find that Slack message, you already forgot what Emily’s email said.

Hit home? Well, as AI models expand to Coworking and Scheduled agents, we have a new best friend that doesn’t really have a name.

(Until we randomly named it. Lolz)

Scheduled Agentic Context Carry. You need to know what it is, why it’s important, and how to use it.

Also on the pod today:

• Scheduled AI agents explained 🗓️
• 1M token context windows 💾 
• Context carry bridges chatbots 🤝 

 

It’ll be worth your 34 minutes:

Listen on our site:

Click to listen

Subscribe and listen on your favorite podcast platform

Listen on:

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

New AI Tool Spotlight – Kollab is an AI-native workspace where your teams and AI Agents seamlessly collaborate, Typewise is an AI Agent Platform for Customer Service, Design.md is a Design system that pulls inspirations from popular websites.

Gemini Embedding 2 — Gemini Embedding 2 is now generally available, bringing stable, natively multimodal embeddings for text, image, audio, and video.

Qwen3.6-27B Released — Qwen3.6-27B squeezes flagship-level coding and multimodal reasoning into a 27B model.

OpenAI Privacy — OpenAI released Privacy Filter, a small local model that detects and redacts PII in long, noisy text with high accuracy.

Atomic Mail — Atomic Mail now runs AI features with Chrome’s Gemini Nano locally, so all message processing stays on your device.

Anthropic Economics Survey — Anthropic is launching a monthly Anthropic Economic Index Survey to gather firsthand accounts of how AI is changing work and expectations.


Anthropic Agents — Production agents scale best with MCP, because a single remote server gives cloud agents standardized auth, discovery, and rich tool semantics.

GPT-5.5 Takes Top Model Spot — On Artificial Analysis, OpenAI’s newly released GPT-5.5 took over the top spot.

Anthropic Study on Jobs — A survey of 81,000 Claude users finds workers in roles where Claude does more tasks are more worried about AI job loss, especially early-career employees

Deepmind Updates — Google DeepMind shows Decoupled DiLoCo keeps large AI training runs alive even when chips and sites fail.

Perplexity Research — Perplexity combines SFT then on-policy RL to boost search accuracy while preserving guardrails.

Applied Digital — Applied Digital has signed a 25-year, $7.5 billion lease to build a large AI data center campus with a U.S. hyperscaler, aiming to supply the surge in compute for generative AI.

ChatGPT and Health — OpenAI rolled out ChatGPT for Clinicians, free to verified U.S. providers, with cited clinical search and HIPAA-ready options. Want faster, evidence-backed answers and less paperwork?

Google Engineer Scandal — Google says a former engineer stole AI secrets and sent them to China, sparking explosive Senate testimony.

1. U.S. accuses China of large-scale AI IP theft 🫣

The White House this week accused Chinese entities of running "industrial-scale" campaigns to extract U.S. frontier AI capabilities, saying they use thousands of proxy accounts and jailbreak techniques to siphon proprietary model knowledge, and announced it will share findings with U.S. companies and consider actions to hold actors accountable.

The memo, disclosed days before President Trump’s planned Beijing visit, raises fresh tensions in the tech rivalry and could complicate decisions about exporting advanced AI chips to China

2. OpenAI and Microsoft tighten AI defenses with deeper cybersecurity partnership 🛡️

OpenAI and Microsoft announced an expanded collaboration to give Microsoft Trusted Access to OpenAI’s most cyber-capable models while Microsoft applies its full cybersecurity team to protect shared models and infrastructure.

The move is timely as AI tools gain offensive and defensive cyber capabilities, and both companies frame the partnership as a way to harden software supply chains and help customers manage emerging risks. Commenters raised governance questions about who decides when models act defensively versus offensively and stressed that model-level protections are only part of a broader security posture.

3. OpenAI unveils GPT-5.5 “Spud” in rapid response to competitors 🥔

OpenAI on Thursday released GPT-5.5, codenamed Spud, a faster and more capable model that handles messy, multi‑step tasks with less prompting and is available now to paid ChatGPT and Codex users while API access is coming soon.

The company says improvements are strongest in coding, office work and early scientific research, and teams reported saving up to 10 hours per week on complex workflows. NVIDIA’s new chips and partnership helped cut inference costs dramatically, a key factor for enterprise adoption as compute becomes central to AI deployment.

4. OpenAI launches workspace agents for teams 🗂️

OpenAI announced that ChatGPT now supports shared workspace agents in research preview for business, enterprise, education, and teachers plans, letting teams create cloud-run agents that handle multi-step workflows while admins control permissions and connected tools.

These agents extend GPTs with Codex-powered abilities to run code, access files and apps, remember context, and operate on schedules or in Slack, aiming to reduce repetitive coordination and speed up team work. Built-in safeguards, admin controls, monitoring via a Compliance API, and analytics are included to manage data access, require approvals for sensitive actions, and review agent performance.

5. Microsoft makes Copilot agentic in Word, Excel, and PowerPoint 📜

Microsoft today made agentic Copilot features generally available in Word, Excel, and PowerPoint, letting the assistant take multi-step, app-native actions directly inside documents, spreadsheets, and slides.

The update shifts Copilot from a passive helper to an active collaborator that can edit, format, and transform files while preserving user control and preferences, and Microsoft says early data shows higher engagement and satisfaction. The company plans further work to expand native actions, improve transparency about changes, and unify the Copilot experience across apps, emphasizing reliability for high-stakes use.

6. Software stocks wobble after ServiceNow, IBM results shake AI-era confidence 📉

Tech markets turned sharply negative Thursday as ServiceNow plunged 17% after citing Middle East conflict headwinds and narrowly beating estimates, while IBM slid 9% despite beating revenue and earnings, fueling fears that AI upends traditional cloud subscription models.

The selloff rippled across the software sector, with Salesforce, HubSpot, Adobe, Intuit, Oracle and Workday all falling and the software ETF down roughly 5% on the day. Investors are pricing in disruption from AI startups like OpenAI and Anthropic that could pressure recurring revenue, even as big tech companies reporting next week have held up better.

Everyone spent 2025 declaring it the year of AI agents. It wasn't.

And 2026 won't be either. But something way more useful quietly took shape across every major lab.

Wait, what?

Anthropic, OpenAI, Microsoft, Google, and Perplexity all shipped scheduled agents this spring. Most leaders kept scrolling.

You know the inner monologue: "Scheduled agent. Sounds dev-only, not for me."

Biiig miss.

This prolly marks the biggest shift in how AI actually works since ChatGPT. We don't say that lightly.

What's the unlock? An AI that wakes up on its own, pulls context across your email, calendar, Slack, drive, and CRM, then hands you finished work before your laptop is open.

Don't worry. We've got your scheduled-agent 101 for the nontechnical leader.

On today's Everyday AI, we name the workflow pattern hiding under every major spring launch, plus the three moves leaders gotta make this week.

You've gotta window. Let's get after it.

1. Name the pattern your rivals already run 🔥

Scheduled Agentic Context Carry sounds like a mouthful until you break it into pieces.

Scheduled means the agent runs on a cadence you pick, not just when you poke it.

Agentic means it calls tools and makes decisions across your stack, while Context Carry keeps memory, preferences, and tool access alive between runs.

Put the pieces together and you get an AI that queries your whole stack without you stitching anything together.

So why is this landing right now?

Three forces collided: million-token context windows, wide app access through MCP connectors, and a scheduling layer that finally reached production.

For 20 years of SaaS, the human has been the duct tape between apps, and that duct tape job is finally over.

Leaders who name SACC first deploy SACC first.

Try This

List the three manual carry tasks you do every morning before coffee, like the inbox triage, the calendar scan, and the industry pulse check.

Those three are your first SACC agents, and they're boring on purpose.

Hand the easiest one to your approved platform this week, run it for five days, then feel what mornings taste like without the duct tape weighing you down.

2. Stop prompting reactively. Start running on a cron ⚡

Most leaders still treat AI like a reactive chatbot, meaning they ask, re-explain, burn 10 minutes, then try again tomorrow.

That playbook is already a year out of date.

So what does SACC actually look like inside your company?

Picture an agent waking every Sunday at 9 PM, pulling fresh industry white papers, scanning your CRM for buyer signal shifts, and flagging a major account that entered your region.

By Monday at 7 AM, a finished brief sits in your inbox, not a chat thread begging for prompts but a deliverable you can forward.

The unit of work shifted on you overnight.

Companies pulling ahead in Q4 aren't writing cleverer prompts. They're running SACC on autopilot while competitors stare at a blank box.

Try This

Pick one recurring task that eats your week, like the Monday status roll-up, the Sunday industry scan, or the Friday pipeline review.

Build it once inside your approved platform, wire the data sources, then run it manually twice to sanity check the output before you trust it.

Then schedule it and walk away. Your Monday morning hook is locked in for good.

3. Work the three-step deploy, in exact order 🚀

Connect, context-stuff, iterate. Each step sets up the next, so the order matters.

Step one, wire live data sources and preferences first, so the agent knows who you are before it touches anything.

Step two, once those connections are live, context-stuff a dedicated memory thread with your docs, goals, and constraints in one place.

Million-token windows now hold weeks of working memory, so load everything into one thread and fork it when the direction shifts.

Step three is where most leaders skip and get burned.

Iterate with chain of thought reasoning before scheduling, because two identical runs can drift wildly by day three on a generative model.

Review the work trail, watch which tools got called, then let it ship.

Try This

Block 45 minutes this Friday, bring your one recurring task, and walk it through all three steps in exact order.

Do not skip iteration, because that stage is where every costly agent mistake gets caught early.

Seven manual dry runs before anything touches a client, no exceptions. By next Friday, one SACC workflow will be shipping on its own while you sleep.

Reply

or to participate.