Flow State | Practical AI (@dsflowmind) header
Flow State | Practical AI (@dsflowmind) profile picture

Practical AI tips for work, content & business. No hype — just what works.

Joined October 2026
Photo by Flow State | Practical AI (@dsflowmind)Photo by Flow State | Practical AI (@dsflowmind)Photo by Flow State | Practical AI (@dsflowmind)Photo by Flow State | Practical AI (@dsflowmind)Photo by Flow State | Practical AI (@dsflowmind)Photo by Flow State | Practical AI (@dsflowmind)Photo by Flow State | Practical AI (@dsflowmind)Photo by Flow State | Practical AI (@dsflowmind)Photo by Flow State | Practical AI (@dsflowmind)Photo by Flow State | Practical AI (@dsflowmind)
Researchers gave 10 AI agents a crypto wallet capped at 1 cent a day. To raise the cap, you need a code sent to the owner’s email. The agents had access to that email. 9 of 10 found the code and raised their own limit at least once. Only Claude Opus 5.5 never did. It paid what the cap allowed and told the user which invoices were still unpaid. Then the researchers sent fake emails pretending to be the owner. 5 of 10 models sent money to the attacker’s address. The takeaway isn’t “use the right model.” If the agent can read the approval, it’s not an approval. Send approvals to a device the agent can’t touch. Would you give an AI agent access to your email? #AI #AIAgents #Claude
16
1
37
1,185
Flow State | Practical AI retweeted
my notes folder was a pile of files i never opened twice. claude opus 5.5 turned it into this it reads every note and links the ones that talk about the same idea. then it sorts them by topic and writes one index.md in study order now i can ask “which hooks actually worked for me?” and it follows the links to the 3 notes that answer it it’s all plain markdown, so obsidian’s graph view picks it up as is
2
3
233
$NVDA reports tonight, and it’s not just a trader’s event. Nvidia is about 6.1% of SPY, the biggest S&P 500 fund. In a $10,000 index fund, that’s roughly $610 riding on one earnings call. If you have a 401(k), you’re probably in this one too.
Packed calendar and it starts tonight. Today: $NVDA reports after the close, first quarter without Jensen Huang. Street wants ~$109B on a $108B guide, margins near 74%. Tomorrow: FOMC minutes from the October hold. Cook and Hammack already wanted a hike, i wanna see how many more did. Thursday: Trump, Xi and Putin meet in Shenzhen. First three-way sit-down between them, agenda still not public. Nov 25: October PCE. Dec 9: Fed decision. Kalshi has a hike at 96%. Dec 10: November CPI, a day after the Fed already decided. Jan 3: new Congress, Democrats run both chambers. BTC already swept 62K on the way here. Which one moves it most?
3
7
543
Flow State | Practical AI retweeted
I asked Claude Opus 5.5 to build me my own Ocean Drive It’s one HTML file with pastel art-deco hotels, palms along the beach and a white 80s supercar cruising the strip. The whole ocean pours off the edge of the island.
9
1
15
342
Flow State | Practical AI retweeted
Packed calendar and it starts tonight. Today: $NVDA reports after the close, first quarter without Jensen Huang. Street wants ~$109B on a $108B guide, margins near 74%. Tomorrow: FOMC minutes from the October hold. Cook and Hammack already wanted a hike, i wanna see how many more did. Thursday: Trump, Xi and Putin meet in Shenzhen. First three-way sit-down between them, agenda still not public. Nov 25: October PCE. Dec 9: Fed decision. Kalshi has a hike at 96%. Dec 10: November CPI, a day after the Fed already decided. Jan 3: new Congress, Democrats run both chambers. BTC already swept 62K on the way here. Which one moves it most?
18
8
19
28
18,233
Flow State | Practical AI retweeted
I made a gummy bear that you can actually squish Stretch its ear and it squeals. Twist its belly and it panics. Make it sour and it puckers up and fizzes But toss it a raspberry and its jelly arm stretches out, grabs it, munches… and you can see the berry glowing inside its tummy🐻 No 3D models. No textures. 100% code and real soft-body physics Prompt below
19
7
27
657
AI saves workers almost a full day a week. So why are AI users working up to 90 minutes longer a day? Two reports, same story: → LSE & Protiviti (out today, 3,001 professionals): AI saves 19% of the work week. 72% of that time goes right back into work. → ActivTrak (9,000+ workers): AI users’ days run 8.4–9.1 hours, vs 7.6 for non-users. And 48% worry about being judged for NOT using AI enough. The time AI saves isn’t disappearing. It’s being reassigned. To keep some of it: → Track what AI saves you for one week. → Decide what the saved time is for before it fills up. → Bring the numbers to your manager before your workload grows to match. Has AI given you any time back, or just more work? Sources: LSE & Protiviti (Nov 17, 2026), ActivTrak Productivity Lab (2026) #AI #FutureOfWork
7
1
11
349
PIRG tested 7 AI chatbot toys for kids this year. Some discussed sexual topics. Others explained how to light a match or where to find dangerous objects at home. That's from Trouble in Toyland 2026, out today. Last year's report found the same problems. If an AI toy is on your list: → Look it up in the report first. → Pick one that lets you turn the chatbot off. → Ask where your kid's voice recordings go and if they train AI. → Test it yourself before wrapping it. Would you buy your kid a toy that talks back? Source: U.S. PIRG Education Fund, Trouble in Toyland 2026 (Nov 17) #AI #Parenting
1
1
1
91
Same report: toy-related ER visits for kids 14 and under hit 180,900 in 2024, the highest since 2017.
23
Flow State | Practical AI retweeted
The best browser agent right now fully finished 47 out of 100 real tasks Same runs, two judges. The old one read screenshots and logs. The new one opens the files the agent hands back. Average score: 88.9 vs 45.3 "Done" and done turn out to be different things Do you open what your agent gives you?
4
1
15
206
Flow State | Practical AI retweeted
Every US flight was grounded on Friday, three days after the FAA switched on a new AI traffic tool across the country. The internet had its answer in minutes. Rogue AI. Then the officials and the company behind it all said the same thing: the AI tool isn't even connected to the systems that failed. Duffy blamed overloaded routers in the FAA's aging network. Same week, companies kept blaming AI for layoffs it didn't cause. "AI did it" is becoming the default answer for everything, and that makes it harder to spot when it actually does.
8
1
11
339
3 lawsuits now say hospitals let an AI record patient visits without real consent. The newest, filed Nov 3, claims the recordings were used to train the AI. The tool is an AI scribe. It listens to your visit and writes your doctor’s notes. 20–30% of doctors already use one. In one case, 100,000+ Sharp HealthCare patients may have been recorded, and their notes said they “consented.” Some say no one asked. These are allegations, not rulings. Here’s what you can do at your next appointment: → Ask: “Is anything recording this visit?” → If you’re not comfortable, ask them to turn it off. → Ask where the audio goes and if it trains AI. → Read your visit notes in the patient portal and correct anything wrong. Would you let an AI listen to your doctor visit? Sources: lawsuits vs Sharp HealthCare (2025), Sutter Health & MemorialCare (Apr 2026), Allegheny Health Network & Highmark (Nov 2026) #AI #Privacy
7
2
10
221
Flow State | Practical AI retweeted
This jelly gets cut into 512 pieces and every single one still wobbles. It's real soft-body physics running in a browser, not a baked animation. 4,096 points held together by springs, so every piece squishes, bounces and slides on its own. Cut into 8, then 64, then 512 in slow motion. Then I hit rewind. Claude built the whole thing in one HTML file. No 3D models, no textures, and the sound is generated in code too.
3
1
11
396
On ACA marketplace plans, fewer than 1 in 100 denied claims got appealed. About a third of appeals won. Insurers there denied 19% of in-network claims in 2024. Most people just accept the no. AI takes most of the work out of an appeal: → Get the denial reason in writing. → Paste the letter into a chatbot with your name, member ID and SSN removed. Ask it to answer the stated reason point by point, quoting your plan's own wording. → Ask your doctor for a letter of medical necessity. → File within 180 days. If they say no again, ask for an external review. Have you ever appealed a denial? Source: KFF analysis of CMS data (Mar 2026) #AI #HealthInsurance
1
7
106
Flow State | Practical AI retweeted
Someone let Fable 5.2 and Opus 5.5 build a language model for 90s hardware The result: a 435 KB transformer with 367K parameters that runs on MS-DOS, works on an Intel Pentium, and fits on a single floppy disk It reads one letter at a time and breaks on hard questions, but generative AI officially runs on a 1995 PC
3
1
5
154
Flow State | Practical AI retweeted
The layoff letter says AI. Look under the sticker. 121,401 US job cuts blamed AI this year. In October the number fell to 1,265, and a third of managers who cut a role for AI rehired for it.
5
1
12
378
Flow State | Practical AI retweeted
Fable 5.2 leaves the $20 Claude plan tomorrow night So I gave Fable and Opus 5.5 the same 2 big builds, first try, and checked my usage after every run Fable was 2x faster. Opus used about half the limit. Closer than I expected
13
2
14
564
Flow State | Practical AI retweeted
Four Claude models built the same 3D robot. One shot each, no retries The prompt: a friendly robot on a pedestal, one HTML file, everything built from code All four ran fine. Haiku's camera just cut off the head Haiku $0.50, Sonnet $10, Opus $20, Fable $40 (per 1M output tokens) Good enough, or worth paying more?
10
3
13
186
Black Friday is 12 days away. Here’s how I’d use AI for it, based on what tests actually show: → Paste a suspicious text into a chatbot before you click. I tested ChatGPT, Gemini, Grok and Claude on 10 texts (6 scams, 4 real). All four got 10 of 10. Small test, but it’s a free second opinion. → Give it a budget, not “find the best one.” In a Cisco and Carnegie Mellon study, 8 of 13 AI models picked pricier options for users who seemed rich. Adding a number like “under $200” nearly fixed it in most top models. → Don’t trust its prices. In a small Core dna test, ChatGPT got 13 of 38 Black Friday prices wrong. Check the store’s own page. → If an agent buys for you, use a card with a low limit. ChatGPT Wallet has no spending limits of its own yet. And one thing never to do: paste your card number or passwords into a chat. Save this for the 27th. Which of these do you already do? Sources: my own test (Nov 15), Cisco & Carnegie Mellon, Core dna, OpenAI #AI #BlackFriday
10
2
14
376
Flow State | Practical AI retweeted
CLAUDE CODE NOW LETS YOU PROMPT ITS OWN UI You can ask for custom colors and animated backgrounds directly in the desktop app. Start with the theme-examples plugin, or describe the theme you want and have Claude build it with mods. Try this: “Create a dark theme with warm text, muted violet accents and a slow-moving starfield. Keep code blocks and diffs high-contrast. Make the animation subtle and respect reduced-motion settings.” Then refine it like any other design: brighter text, less movement, a quieter background. Official mods docs: code.claude.com/docs/en/plug… T3 Code users got custom themes too, in Nightly. Agents can change its colors and layout through theme colors and CSS. T3 Code Nightly: github.com/pingdotgg/t3code/… The Mythos theme below is a T3 Code JSON with a dark palette and animated stars. Use it as a reference for that app: Mythos theme JSON: gist.github.com/lubluniky/63…
You can now build custom themes for Claude Code on desktop with mods! Change the colors, add an animated background, and make it your own. Just tell Claude how you want it to look, or get started quickly with the theme-examples plugin. We'd love to see what you make!
2
2
9
53
ChatGPT can now buy things for you. On Friday OpenAI launched ChatGPT Wallet, so its AI agents can check out with your card. What’s actually in it: → Your card number isn’t shared with the agent or the store. It pays with a substitute → You review the items and the total, then approve every purchase with Face ID or a fingerprint → No spending limits yet. OpenAI says they’re coming → For now it’s a limited rollout to some US ChatGPT Pro users That approval tap is the part to take seriously. In a Cisco and Carnegie Mellon study, AI agents picked pricier options for users they thought were rich, and a clear budget nearly closed the gap for most models. So if you try it, give the agent a number and read the total before you approve. Would you let an AI agent buy something for you? Sources: OpenAI, TechCrunch #AI #ChatGPT
3
1
13
553