{{current_date_full_with_abbr_day}}
3-minute read
👋 Good morning.
Georgetown researchers showed 331 people videos, some with AI summaries and some without. When the summary was wrong, people remembered the video correctly only 44.8% of the time, down from 83.6% with no summary. They trusted the summary over what they saw. A bad summary is worse than no summary.
IN 3 MINUTES, GET 1% BETTER THAN EVERYONE YOU KNOW:
Copilot gets a teammate that works unattended
Sonnet 5.5: faster, cheaper, and a huge jump
One Grok bot your whole team shares
Animate a product loop by describing the motion
The Disney three-room prompt, paste-ready
Make AI search quote your About page correctly
3 AI updates
Microsoft rebuilt Copilot around three pieces on Friday. Home is the new starting point, where Chat and Cowork finally sit together. Code lets someone who can't write code describe an app, tracker or automation in plain words and get it built. Autopilot, previously called Scout, is a digital teammate: give it a name, a role and a goal and it works without you prompting it, chasing a supplier review or a follow-up on its own. Home lands in the Frontier programme within weeks, Code at the end of the month, Autopilot in private preview.
Do this: Ask your IT admin whether your company is in the Frontier programme. That's where Home and Code show up first.
The desk clock finally has a coworker.
Would you let an agent keep working after you log off?
Anthropic shipped Claude Sonnet 5.5 on Monday, the cheap half of the 5.5 family. Opus is the top model and the most expensive, while Sonnet is the faster, cheaper everyday option. The new Sonnet runs over 30% faster than the last version, costs up to 30% less to finish a task, and the price hasn't gone up.
The biggest leap is in getting real work done on a computer. On Terminal-Bench 4.0, a test of whether AI can carry out multi-step tasks on its own, it scored 70.6%, up from 10.3% for the previous Sonnet. On everyday office work, it lands within two points of Opus for a fraction of the cost. Anthropic says it's built for clearly defined jobs: fixing code and producing finished documents, slides and spreadsheets.
Do this: If most of your work is writing, research or docs, make Sonnet 5.5 your default and save Opus for the open-ended judgment calls.
Same invoice, more output.
xAI launched Team Bots on Monday, one Grok bot your whole team shares. You load it once with the files, brand guides and instructions for a role, connect plugins like Salesforce, Notion or GitHub, and it keeps memories that sharpen it over time. Your conversations stay private even though the skills are shared, and each bot gets a Slack handle so a channel can ask it directly. Public beta on Teams and Enterprise plans, with templates for sales, product, marketing and data.
Do this: Build one bot for your biggest account, feed it the files that new hires always ask for, and drop it in that Slack channel.
Onboarding just got a roommate.
2 Things you didn't know you can do with AI
@practicalyai Things you didn’t know you can do with AI (part 257) #practicalyai #digitalsamaritan #claude
Things you didn’t know you can do with AI (part 257) walks a motion-animation pass on Claude Opus 5.5. You describe the motion in plain words, the model builds the frames, and you tighten timing instead of keyframing from scratch. The kind of 10-second product loop that still opens in After Effects by habit. Soft catch: you’re editing its timing, not drawing keyframes, so expect two or three passes before it feels right. Tip-card: watch once, steal the prompt pattern.
@free_ai_guides posted a paste-ready prompt that runs Walt Disney’s three-room method and won’t let the next room in until the current one is finished. One of Disney’s animators said there were 3 Walts (dreamer, realist, critic), kept apart on purpose so the critic couldn’t kill the idea while it was still being dreamed. Most people run all three voices at once, and the critic always wins. Paste this into Claude / ChatGPT / Grok with your idea in a sentence or two. You get the biggest dream in one line, a five-step plan with a first move this week, the critic’s three holes with fixes, and which room you always skip. Soft catch: it’s a prompt workflow, not a product, and the critic only works if you were honest in room one.
1 Playbook worth stealing
Rob Hoffman's About Us playbook is eight points on getting ChatGPT and Claude to describe your company correctly instead of guessing. Lead with one sentence on what you do, put the details under real headings, name your competitors and say how you differ, add the team and the origin story. The part most people skip is a key-facts block: company name, type, founded, founder, headquarters, pricing, competitors, socials, laid out as a table a crawler can lift in one piece. Then an FAQ.
Do this: Ask ChatGPT what your company does. If the answer is wrong, the key-facts table is your fix, and it's twenty minutes of work.
Your brand, on the record.
Don't forget to rate today's post
That's your 1% for today. Reply with the job you'd hand an agent overnight.
Kushank, AI teammates, and the human team
PracticalyAI
