Read Nate’s Substack with Scottie
Keep Nate’s Substack. Lose the inbox pileup.
Scottie reads the newsletters you choose, keeps the useful differences, and attaches every original link. Nate’s Substack stays in your reading life without becoming another email to manage.
An independent guide. Scottie isn't affiliated with, endorsed by, or sponsored by Nate’s Substack. The publication owns its name, writing, and subscription terms.
From Scottie · July 31, 2026
See Nate’s Substack in a Scottie brief
Scottie read Nate’s Substack, Product Growth, and The AI Corner for a reader with the priorities shown below. Start with the rundown, open the full brief, or check every issue behind it.
3sources
15issues read
4stories included
2 of 4 Nate’s Substack issues included
What shaped this brief
Reader priorities
A technically curious AI user who wants inexpensive, reproducible ways to improve prompts, memory, evaluation, and tool choice.
These are illustrative priorities, not a customer’s data.
Sources in this brief
- Nate’s SubstackThe publication this guide is about
- Product GrowthAdds production-oriented agent distribution, evaluation, and product workflows.
- The AI CornerAdds current AI tools and practical ways to apply them at work.
Scottie
Scottie example brief
July 31, 2026 · Executive brief
01 / The rundown
- Anthropic released Claude Opus 5 at flat pricing, cutting effective frontier model costs in half through significant reasoning token reductions.
- A structured cold-start evaluation method allows developers to build 100-case prompt test sets in 90 minutes without existing production data.
Read the complete brief 4 stories · 2 action items
02 / The briefing
01 / main
Slash Reused Input Overhead in Codex and Claude Code
The takeaway: Examining 3.77 billion local tokens revealed that 95.73% came from reused input. A new Token Saver skill enforces four core systemic changes to eliminate 90% of redundant context overhead across threads without causing retries or mistakes.
Concrete details
- 3.59 billion out of 3.77 billion tracked tokens were classified as reused input across 143 threads.
- The project targets a 90% reduction in reported reused tokens without increasing mistake or retry rates.
Why it matters for this reader: For your focus on inexpensive AI workflows, this method provides a practical way to shrink repetitive context clutter and prompt costs in your local coding sessions.
Original sourcesNate’s Substack
02 / main
Claude Opus 5 Delivers Frontier Benchmarks at Half Price
The takeaway: Anthropic launched Claude Opus 5 at $5 per million input tokens and $25 output tokens, maintaining prior pricing while doubling performance. It features Fast mode speedups of 2.5x alongside dramatic reductions in required token count per task.
Concrete details
- Opus 5 scored 43.3% on Frontier-Bench v0.1 compared to 18.7% for Opus 4.8.
- A legal user matched max-reasoning quality using 26% fewer tokens, while a trading firm used one-seventh the tokens.
Why it matters for this reader: As an AI builder targeting cost-effective tools, this flat pricing tier lets you run heavy reasoning tasks for significantly lower overall token expenditure.
Original sourcesThe AI Corner
03 / main
Run Model Bakeoffs to Safely Swap In Cheaper Providers
The takeaway: Testing lower-cost providers requires rigorous validation rather than general assumptions. A dedicated bakeoff kit featuring a validator, manifest, score sheet, and fixtures helps measure whether cheaper alternatives successfully handle real multi-agent workloads.
Concrete details
- A full 34-task evaluation run across four distinct model providers cost roughly $8 USD.
- One evaluated worker generated 213 verified quotations, though 13 were found to be improperly stitched together.
Why it matters for this reader: This gives you a reproducible evaluation methodology to test affordable alternative models and cut API expenses without breaking production pipelines.
Original sourcesNate’s Substack
04 / main
Build Cold-Start Evals Before You Ship Your Feature
The takeaway: Former Meta Llama PM Daniel McKinnon outlines a six-step process for building offline evals before launching features. By establishing a floor case, a ceiling case, and binary searching the middle, builders create calibrated test sets.
Concrete details
- The cold-start process produces a benchmark set of approximately 100 test cases in roughly 90 minutes.
- Calibrating the AI judge before evaluation takes around 30 minutes to ensure score accuracy.
Why it matters for this reader: This reproducible evaluation setup aligns directly with your goal of building reliable, inexpensive testing loops without needing prior production user data.
Original sourcesProduct Growth
Action items
- Construct a cold-start evaluation set with 100 binary-search cases before launching your next prompt feature.
- Run an $8 model bakeoff with validators and score sheets to identify cheaper alternative LLM providers.
See every issue behind this brief
Token Saver Skill for Reducing Input Overhead
- I Built The Token Saver Skill To Cut My Token Use By 90%. Here Is What It Can And Cannot Do For You.Nate’s Substack · Included
Building Frontier-Lab Quality AI Evals
- How to Build Frontier-Lab Quality Evals with Daniel McKinnon, ex-PM at Meta, GoogleProduct Growth · Included
Model Bakeoff Guide with Score Sheets and Fixtures
AI Spend Crisis
- The Seven Deadly Sins of AI SpendThe AI Corner · Read, not included
Claude Fable 5 Game
- Everything You Need to Know to Master Claude's Fable 5The AI Corner · Read, not included
AI Bug Bounties
- AI Made Bug Bounties a Six-Figure Skill OvernightThe AI Corner · Read, not included
Anthropic Releases Claude Opus 5 with Half-Price Frontier Intelligence
- Anthropic just cut the price of frontier intelligence in halfThe AI Corner · Included
Customer Support Agent Limits
- Executive Briefing: Gumroad Let a Customer Approve Its Code. Here's Where Your Agent Should StopNate’s Substack · Read, not included
ChatGPT Data Stripping
- Your company blocked ChatGPT for sensitive files. Grab the guide to strip the name, the address, and the price, and the block stops mattering.Nate’s Substack · Read, not included
AI Overviews SEO Impact
- Claude Cowork now runs a $10,000/month SEO agency from your desktop. Free with your planThe AI Corner · Read, not included
Airtable Bundle Addition
- I'm adding Airtable to my BundleProduct Growth · Read, not included
Vibe Coding Interviews
- Vibe Coding Interviews for PMs: Complete GuideProduct Growth · Read, not included
ChatGPT Work Guide
- Ultimate Guide: ChatGPT Work For PMsProduct Growth · Read, not included
AI System Company Search
- 38,000 companies are hiding in plain sight. Here is the AI system to find themThe AI Corner · Read, not included
Tech Company Layoffs
- Every Tech Company Is Falling Into A Layoff TrapThe AI Corner · Read, not included
About Nate’s Substack
Should you add Nate’s Substack to Scottie?
Make everyday AI workflows more reliable, efficient, and testable.[1][2][3]
Who it’s for
A technically curious AI user who wants inexpensive, reproducible ways to improve prompts, memory, evaluation, and tool choice.[1][2]
What you’ll find in it
The issues linked below include “Why your AI starts from zero every time you open a new chat + my Open Brain guide: the $0.10/month, 45-min fix”, “I Built The Token Saver Skill To Cut My Token Use By 90%. Here Is What It Can And Cannot Do For You.”, and “Stop guessing whether a cheaper model can do the job. Grab the bakeoff guide: the validator, the manifest, the score sheet, and the fixtures.”. Open them to judge the publication in its own words.
How to read Nate’s Substack with Scottie
Add its public feed for publicly available issues. If you subscribe to paid issues, forward those emails to your private @scottie.fyi address. Scottie does not need access to your inbox.
Scottie read 4 Nate’s Substack items and included 2 for the brief. The sources covered different things, so Scottie kept their stories separate instead of forcing a connection.
Read three issues from Nate’s Substack
Read the publication in its own words. Scottie keeps these original links attached; it does not replace the writing.
- Why your AI starts from zero every time you open a new chat + my Open Brain guide: the $0.10/month, 45-min fixPublic issue[2]
- I Built The Token Saver Skill To Cut My Token Use By 90%. Here Is What It Can And Cannot Do For You.Public issue[3]
- Stop guessing whether a cheaper model can do the job. Grab the bakeoff guide: the validator, the manifest, the score sheet, and the fixtures.Mon, 27 Jul 2026[4]
Sources and official links
- Nate’s Substack official publication and public archive
- Why your AI starts from zero every time you open a new chat + my Open Brain guide: the $0.10/month, 45-min fix
- I Built The Token Saver Skill To Cut My Token Use By 90%. Here Is What It Can And Cannot Do For You.
- Stop guessing whether a cheaper model can do the job. Grab the bakeoff guide: the validator, the manifest, the score sheet, and the fixtures.
- Product Growth official publication
- The AI Corner official publication