Claude adds protein design to its resume

PLUS: Use the Loop Method for better ChatGPT results

Zach Mink

August 20, 2026

Good morning, AI enthusiasts, and welcome to our 12,567 new readers. Four days ago, Dario Amodei went on X to say Anthropic's biology work was months away from its first "early glimmers." Did he undersell the timeline?

Claude just added protein design to its already lengthy resume, with new research showing the models running an early step of the drug discovery pipeline on their own, and producing results that held up in the lab.

Reminder: Our next live workshop is today at 12 PM EST — Join and learn how to land your first paid AI consulting gig on Upwork. RSVP here.

In today’s AI rundown:

Anthropic's Claude tackles protein design
Nate's Notebook: Uber's tiger team playbook
Use the Loop Method for better ChatGPT results
Replit gives routine work a cheaper route with Free Mode

LATEST DEVELOPMENTS

ANTHROPIC

🧬 Anthropic's Claude tackles protein design

Image source: Anthropic

The Rundown: Anthropic just published new research showing Claude ran protein-design campaigns largely on its own, a key early step in drug discovery, with its designs yielding working molecules on 14 targets out of 15 at success rates above the industry norm.

The details:

The company tested its Mythos Preview and Opus 4.8 models, letting them run autonomously with one expert-written prompt, internet access, and tools.
Anthropic didn’t do the actual lab work, with Twist Bioscience and Adaptyv Bio creating the candidates in their own labs and running the measurements.
The Claude models hit 22-35% success rates on molecules that actually gripped their target, which Anthropic says beats the typical 10-15%.
Opus 5 separately opened raw instrument files with no lab software, measuring a sample at 96.4% pure in 19 minutes; the lab’s own report took four days.

Why it matters: CEO Dario Amodei said Anthropic hoped for “early glimmers in the coming months” in biology and medicine in an X post last week, and just days later we have a spark. While AI in protein design isn’t new, the difference here is that a general model got these impressive results while running the campaign itself.

TOGETHER WITH UNWRAP

🚒 See what's burning before it becomes your next fire drill

The Rundown: Somewhere in your customer feedback right now, a small issue is about to become a big one. Unwrap gives you AI-powered visibility to catch issues while they’re still embers, not after they’ve become a fire.

Unwrap’s proactive approach gets you:

50% faster root cause investigation on support tickets
24+ hours saved across teams per month
4x fewer ticket spikes
One view of customer sentiment across every channel

Rundown readers get free access with a zero cost, zero commitment trial. See what's burning before it becomes your next fire drill!

NATE’S NOTEBOOK

🐯 Nate's Notebook: Uber's tiger team playbook

The Rundown: Each week, Rundown AI educator Nate Grahek breaks down what he's seeing while teaching real teams to use AI. This week he kicks off a three-part series on capturing AI ROI: big companies, small businesses, then individuals.

Nate: Remember when Uber made headlines for blowing through its entire 2026 AI budget in four months? Their CTO started capping employee AI use and declared the "tokenmaxxing era" over.

Now the same CTO, Praveen Neppalli Naga, has posted what Uber did next. That's what makes this case study worth your attention: it's a company learning in public from an expensive mistake.

I've been saying it for months: unsubsidized, full-rate frontier models are simply too expensive to hand to everybody in a large org and let them figure it out. And enterprise AI is harder to begin with, having to navigate security, large data complexity, and privacy.

Uber's answer was the scalpel. They call them Agentic Pods, and each is only two people: one AI-proficient engineer and one domain expert from Finance, Marketing, Ops, etc. running a surprisingly short 10-day sprint. Shadow the expert, build alongside them, ship.

The result: a financial pacing report that took a key player two full days now runs them 10 minutes. That's the mindset I want big companies to steal. Your first AI projects have to earn you the right to swing bigger. Enterprise needs to start acting more like bootstrappers: quick, pragmatic efficiency wins, not the flashy long-roadmap bet.

Next week, part 2: How small businesses can take advantage of this moment while many enterprises are too slow to figure this out.

AI TRAINING

🔂 Use the Loop Method for better ChatGPT results

The Rundown: Learn how to improve any repeated ChatGPT workflow with the "Loop Method." We created it to improve our video editing and image generation skills, but you can try it on anything!

Step-by-step:

Open ChatGPT Work and find a workflow that needs improvement: a skill, project folder, or process that works inconsistently.
Ask it to run three loops with a review panel. Use this template: "Improve this workflow in 3 loops. Have a panel of sub-agents adversarially review each loop. Done when: [definition of done]."
Use /goal to assign ChatGPT a deliverable or capability you want the workflow to build.
Let ChatGPT run the three loops. It could take anywhere from 15 minutes to an hour. Lower-thinking models like Luna will be faster!
Test the improved workflow on a second example. If it still needs improving, tell Codex to codify ambiguous work into scripts.

Going further: Turn this self-improvement process into a reusable skill, then run it whenever a workflow becomes inconsistent.

PRESENTED BY AWS

🏟️ Join 60,000 builders and leaders at AWS re:Invent

The Rundown: AWS re:Invent brings 2,200+ sessions to Las Vegas, and 70% are hands-on. Join builders’ sessions, workshops, code talks and more to learn by doing, not just through sessions.

While onsite, you can:

Bring edge-case questions to engineers and partners building on AWS
Test services against realistic workloads in sandbox environments
Access certification exam discounts, with bootcamp-style prep during the week

REPLIT & OPENAI

💸 Replit gives routine work a cheaper route with Free Mode

Image source: Replit

The Rundown: Replit just launched Free Mode, a new default setting for its $20 and $100 paid plans that runs everyday chats and tasks on OpenAI’s GPT-5.6 Luna without touching a user’s credit balance — with the company crediting the Luna’s 80% price cut in July for making the math work.

The details:

Replit says its $20/mo Core plan scales as much as 30x usage with Free Mode, adding 30 hours per month of chat using the efficient model.
Bigger builds still burn credits in the higher-performance modes, and the agent can route to OAI's Sol for a harder step before handing users back to Luna.
Replit president Michele Catasta called the default-on setup “radical”, arguing plenty of tasks need a smaller, cheaper model, not an expensive frontier one.
Catasta said Free Mode is "the first of many launches" with OAI, with Sam Altman saying he foresees a "renaissance-level entrepreneurial boom."

Why it matters: Everyone went from tokenmaxxing to counting tokens, and Replit is turning Luna's speed-cost-intelligence balance into pure user value without the previously expected budget-model drop-off attached. When low prices stop meaning low intelligence, the economics of a whole world of products start to change.

QUICK HITS

COMMUNITY AI WORKFLOW OF THE DAY

▸ Matthew built a weekly safety-course generator for his son

Today’s workflow comes from reader Matthew M.:

“I wanted a way to teach my son practical safety and life skills, starting with how to swim confidently and what to do if he gets into trouble in the water. I needed something more engaging than simply talking at him, but writing a polished, interactive lesson from scratch every week was not sustainable.

First, Claude and I designed a single interactive HTML course as a proof of concept. It was a swim-safety course with a branded look, including a custom color palette, fonts, and a progress tracker styled like pool lanes.

I wanted to reuse the same format for a new topic every week. Claude documented the entire course format as a standing "brief" document. Each week, I open Claude Cowork, paste in the brief and that week's topic, and Cowork returns a finished, self-contained HTML file.”

See Matthew’s workflow here at The Rundown University. How do you use AI? Tell us for a chance to be featured.

🛠️ Trending AI Tools

🔒 Incogni - remove your personal data from the web so scammers and identity thieves can’t access it. Use code RUNDOWN to get 55% off.*
📚 Student Hub - Google’s Gemini-dedicated area for students, with study notebooks, quizzes, flashcards and more
🚀 Replit Free Mode - Replit’s new Luna-powered credit-free usage mode
🤖 GLM-5.3 - Z AI’s open model with strong coding, agentic, and cyber skills

*Sponsored Listing

📰 Everything else in AI today

Glean:GO San Francisco, August 26-27 - The intelligence era is here. Join us in person or virtually to make AI part of your everyday work and get more time back for the work that matters. Register now.*

Stripe officially announced its acquisition of OpenRouter, while reportedly telling investors that Jan. 1, 2026 was the start of “the singularity”.

Chinese AI lab Z AI’s recently released GLM-5.3 officially scored a 60 on Artificial Analysis’ Intelligence Index, moving into fourth place overall and tying Kimi K3 for the top open model.

U.S. AI chipmaker Cerebras introduced CS-4, its fourth-gen AI computer, which it says responds at up to 30x the speed of GPU-based rivals even on the industry’s largest models.

OpenAI previewed Private Safety Processing, a system that scans for misuse across a user's API sessions without staff ever seeing the content, letting it keep its zero-data-retention promise on frontier models.

*Sponsored Listing