- Everyday AI
- Posts
- Ep 786: 2026 LLM Cheat Code: 10 Essential Steps To Get the Most out of Any AI Chatbot (Start Here Series Vol 26)
Ep 786: 2026 LLM Cheat Code: 10 Essential Steps To Get the Most out of Any AI Chatbot (Start Here Series Vol 26)
Report: Microsoft building new model, Apple's AI play leaks, Musk downplays Anthropic compute deal, Claude 4.8 model around the corner and more.
Everyday AI
May 28, 2026
👉 Subscribe Here | 🗣 Hire Us To Speak | 🤝 Partner with Us | 🤖 Grow with GenAI
Sup y’all 👋
I get messages often about asking me to create a paid-tier for more/better access, like this from Adam this morning. (Thanks for letting me share, Adam!)
I always want to keep Everyday AI as the best free source of AI info for as long as I can, and not flood the airwaves with ads from AI startups that could die in months.
Eventually, I may create some kinda paid tier for people who want more/better access. I’ve always been on the fence.
Thoughts?
What would you (or your company) pay for when it comes to AI enablement?🗳️ Vote to see LIVE results 🗳️ |
|
✌️
Jordan
Outsmart The Future
Today in Everyday AI
8 minute read
🎙 Daily Podcast Episode: We break down the 10 essential steps to getting better outputs from tools like ChatGPT, Claude, Gemini, and more. Give today’s show a watch/read/listen.
🕵️‍♂️ Fresh Finds: GPT-5.6 shows up in Codex logs, Claude testing new web-based Tasks feature, and OpenAI is already preparing safeguards for the 2026 elections, and more. Read on for Fresh Finds.
đź—ž Byte Sized Daily AI News: Report: Microsoft building new model, Apple's AI play leaks, Musk downplays Anthropic compute deal, Claude 4.8 model around the corner and more. Read on for Byte Sized News.
đź’Ş Leverage AI: Big AI companies now just copy each others features, which means we can finally release a set of concrete best practices for getting the most out of any AI chatbot. Keep reading for that!
↩️ Don’t miss out: Miss our last newsletter? We covered: Robinhood lets agents control your investing, Microsoft's big splash in AI images, Google's next enterprise AI play and more. Check it here!
Ep 786: 2026 LLM Cheat Code: 10 Essential Steps To Get the Most out of Any AI Chatbot
This is the Everyday AI episode we probably shoulda done a while ago.... 👇
Because as different as ChatGPT, Gemini, Claude and others actually are under the hood, they have really started to copycat each other over the past 6 months.
Which means we finally have a set of concrete best practices to get the best outputs from any LLM.
Join us as we boil thousands of hours of experience into a 30-ish minute crash course that you can't afford to skip out on.
Also on the pod today:
• AI models copycat trend 🤖
• Free AI plans = risk 🚫
• Desktop apps surpassing web 💻
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 – Pancake is an OpenClaw tool for Slack, Pitch Agent creates on-brand presentations in seconds, Robinhood Agentic Trading lets AI manage your portfolio
Runway MCP — Runway just dropped MCP, so you can generate images and video from tools like Claude, ChatGPT, and Cursor without leaving your workflow.
Microsoft Tightening Intimate AI — Microsoft is tightening its response to non-consensual intimate imagery, including AI-generated abuse, with easier reporting and broader use of StopNCII.org hashes across services.
AI and Media — Amazon MGM is putting real money behind AI-made Prime Video shows
NYT and AI — See why the New York Times is pushing back on AI use.
Claude Updates — Claude’s latest update is all about giving agents more control, with self-hosted sandboxes and private MCP tunnels now available for enterprise use.
Claude Tasks — Claude appears to be testing an online-version of Cowork called Tasks.
OpenAI Funding — OpenAI Foundation is putting $250M into grants, partnerships, and direct work to help people and institutions handle AI-driven economic change.
Erin Brockovich AI Map — Most of the community reports on Brockovich’s AI data center map are coming from Texas, with Sulfur Springs alone accounting for nearly half the state’s total.
OpenAI and Anthropic — AI’s top labs are sending mixed signals on jobs, with Anthropic warning of major labor displacement while OpenAI says the apocalypse talk is overblown.
New OpenAI Model? — A Codex trace log showed a glimpse of gpt-5.6, so OpenAI could have a new model on deck.
Sesame Preview — Sesame just put its iOS app in preview, with four personal voice agents that can browse the web, set reminders, and remember things.
Ozzy Osbourne — Ozzy Osbourne may be back as an AI hologram that can answer fan questions in his own voice.
Youtube AI Label — YouTube is making AI labels harder to miss, with new placement on videos and Shorts.
OpenAI Election — OpenAI says it’s gearing up for 2026 elections with live results, voting info, provenance tools, and tighter misuse detection.
1. Amazon licenses AI shopping tech to retailers 🤖
Amazon is moving its Alexa for Shopping technology beyond its own store and licensing it to other retailers, a timely push to make its AI tools a core part of online shopping across the web.
The company says retailers can launch their own branded AI shopping assistants in as little as 60 days, with Kate Spade already on board and others in testing.
2. Claude Desktop leak teases Opus 4.8 rollout today 🆕
A newly discovered leak from Claude Desktop has put Anthropic back in the spotlight, with code in the app’s model selector explicitly naming “opus-4-8” and tying it to higher reasoning tiers, a sign the model may be staged for release.
The screenshoted logic suggests Opus 4.8 is being prepared for more advanced tasks, especially the kinds of complex coding and agent-style work that rely on stronger reasoning settings. That lines up with a growing trail of recent references to Opus 4.8 and Sonnet 4.8 across source maps, Vertex AI, and internal builds, which has fueled speculation that a launch could be close.
3. Meta rolls out paid AI appsđź’¸
Meta announced Wednesday that it is expanding paid subscriptions worldwide for Instagram, Facebook, and WhatsApp while also testing new plans for creators, businesses, and Meta AI users under a new “Meta One” label.
The consumer tiers add extra features and personalization for power users, while the AI plans split casual use from heavier, more compute-heavy usage, which is a clear sign Meta wants more revenue streams beyond ads.
4. Musk says SpaceX’s Anthropic Colossus lease is 180 days, not years 🛰️
Elon Musk said Thursday that SpaceX’s AI compute lease with Anthropic at its Colossus data center is only a 180-day deal with a 90-day mutual cancellation option, pushing back on the longer timeline suggested in SpaceX’s IPO filing.
That filing said the agreement could run through 2029, which would imply a much bigger revenue stream, but Musk said SpaceX asked for the shorter term and will give Anthropic a reasonable off-ramp if needed.
5. OpenAI Rolls Out Secure MCP Tunnels for Private Servers 🖥️
OpenAI is pushing a timely new way for private MCP servers to connect with ChatGPT, Codex, and the Responses API without exposing anything to the public internet or opening inbound firewall ports.
The setup uses an outbound-only tunnel-client that lives inside the customer network, polls OpenAI for queued work, forwards requests locally, and sends responses back through the same secure path. In simple terms, it lets private tools stay private while still being reachable by supported OpenAI products, with admin controls, health checks, and workspace permissions built in.
6. Microsoft eyes new AI coding model ahead of Build 2026 🏗️
Microsoft is expected to unveil a new AI coding model next week, timed to line up with its Build 2026 developer conference on June 2 to 3, according to The Information.
Details are still scarce, but the timing suggests Microsoft is using Build to push deeper into AI-assisted software development after last year’s Copilot agent and Azure multi-model expansion. The bigger picture is simple: Microsoft is leaning hard on GitHub, Visual Studio Code, and Copilot to keep developers inside its ecosystem while rivals race to improve their own coding assistants.
7. Leaks: Apple’s iOS 27 Siri overhaul lands at WWDC with a deeper AI push 🍎
According to Bloomberg, Apple is set to unveil a redesigned Siri at WWDC on June 8, and the timing matters because the company is clearly trying to turn its assistant into a much more central part of the iPhone experience.
The new version reportedly moves Siri into the Dynamic Island, adds a “Search or Ask” interface, and brings a chatbot-style experience that can tap Apple services as well as outside AI tools like ChatGPT, Gemini, and Claude. Apple is also testing major AI upgrades for the Camera, Photos, and Shortcuts apps, which suggests this is not just a Siri refresh but a wider attempt to make iOS feel more conversational and more useful.
Yeah, really.
The biggest AI tools are starting to blur together. ChatGPT, Claude, Gemini, Copilot, Perplexity, Grok, Codex, Google Antigravity, Cursor.
Different engines, different quirks, same bigger truth: the advantage is shifting from picking the perfect model to building the cleanest operating rules around whichever model your people use every day.
That should scare every enterprise leader trying to scale AI without creating chaos.
Most teams are still winging it with free plans, random prompts, mystery permissions, half-connected files, and one brave person pretending Slack reminders count as governance. Cute.
The opportunity is bigger than “learn better prompts”: choose the right AI operating system, feed it the right context, and inspect the work before agents touch revenue.
That’s what we tackled on today’s Everyday AI show: the LLM cheat code for getting better outputs from any AI chatbot, why the copycat era of AI tools is actually useful, and how to turn the fire hose into a system your company can run without drowning.
1. Stop worshipping the model 🔥
The model race still matters, but obsessing over it is becoming a managerial distraction.
When ChatGPT, Claude, Gemini, Copilot, Perplexity, Grok, Codex, Google Antigravity, and Cursor start acting like different dialects of the same language, your real question changes. Which AI operating system owns most of your team’s daily knowledge work?
That choice matters because context, permissions, memory, files, apps, chat history, skills, plugins, and workflows compound over time. If every employee is building their own tiny AI island, the company isn’t becoming AI-native.
It’s becoming harder to manage.
Also, stop using free or non-thinking models for real business decisions. If the work matters, use the paid account, the right model, and the surface that fits the job, because web, desktop, browser access, local files, and agentic tools all give the system different powers.
Try This
Pick one department workflow and declare the default AI operating system for that work.
Then define the account tier, model type, surface, approved use cases, banned use cases, and escalation path before people start improvising with company data.
2. Feed it before asking ⚡
Bad AI output is often a context crime with a clean UI.
Treat LLMs like generative reasoning systems that pull from whatever they can see: old training data, the web, memory, files, apps, connectors, chat history, and the instructions your team bothered to provide.
That last part is usually where the whole thing goes sideways.
The hidden lesson for leaders is that enterprise AI quality comes from controlling the context layer before anyone asks for an output. Otherwise, you get confident mush that sounds helpful, feels polished, and quietly ignores the business reality sitting inside your docs, CRM, inbox, SOPs, and customer history.
This is where prime, prompt, polish actually earns its keep. Same with RefineQ, 5-5-5, role, goal, sources, constraints, examples, questions, and all the unsexy structure that separates useful work from AI confetti.
Try This
Create a context packet for one recurring executive task.
Include the goal, audience, approved sources, banned sources, strong examples, weak examples, business constraints, output format, and the questions the model must ask before producing anything final.
3. Own the middle steps 🚀
Agents are where productivity gets real, and where lazy governance gets expensive.
Once AI can use your browser, read local files, connect to Google Drive, SharePoint, OneDrive, Box, HubSpot, ClickUp, and your CRM, the job changes from asking a chatbot for help to supervising a system that moves business context across your company.
That’s useful.
Also terrifying, if nobody can see what happened.
Shadow AI with copy-pasted docs was already messy. Shadow agents with read/write access can email customers, update records, move data between tools, and make one tiny assumption at machine speed across a workflow your team barely understands.
So governance can’t be a slide deck that everyone forgets after onboarding. It has to be permission design, observability, reasoning artifacts, verification, iteration, and expert-driven loops that let leaders inspect the steps between the input and the final work.
Try This
Before turning any AI workflow into a skill, plugin, or scheduled agent, run it manually three times with an expert watching the middle.
Capture the prompts, files, connectors, tool calls, approvals, failure points, edits, and final output. Then automate only the version someone can explain without hand-waving.
Reply