New issue every weekday · 7am

The AI briefing
for people who'd
rather build than
doomscroll.

Every weekday morning, Inference turns the firehose of AI news into a five-minute read. No hype, no fear — just what actually shipped and why it matters to you.

A
M
K
+
Join 182,000+ readers. Free forever, unsubscribe anytime.
Inference No. 248
Tuesday·May 27·5 min read
The model that finally nails long-horizon tasks

A new release can hold a goal across hours of autonomous work without losing the plot. We break down what changed, what it means for agents, and the one limitation nobody's mentioning.

  • Two more frontier labs open-weight their mid-tier models
  • A 70% drop in inference pricing reshapes the startup math
  • The eval everyone trusted just got quietly gamed
Driftwood — context that follows you
We tested it for a week. Genuinely useful for anyone juggling more than three projects at once.
Read the full issue → Delivered 7:00 AM
No.247The eval problem nobody wants to talk about No.246Open weights just changed the math for startups No.245Agents are getting good — here's what breaks first No.244The quiet death of the prompt engineer No.243Why your RAG pipeline is probably overbuilt No.247The eval problem nobody wants to talk about No.246Open weights just changed the math for startups No.245Agents are getting good — here's what breaks first No.244The quiet death of the prompt engineer No.243Why your RAG pipeline is probably overbuilt
Read every morning by builders at
Every Issue

Five sections.
Five minutes.

The same dependable structure every day — so you always know where to look, and you can stop reading the second you've got what you need.

[ 01 ]
The Signal
The one story that genuinely matters today, explained from scratch — including why it matters and who should care.
[ 02 ]
Quick Hits
Four or five updates worth knowing, each in a single sentence. Skim it in sixty seconds and move on with your day.
[ 03 ]
Tool Drop
One genuinely useful AI tool, actually tested by a human — not a sponsored link, not a press release. Just what works.
[ 04 ]
Paper Trail
A notable research paper translated out of academ-ese and into plain English you can use in a meeting tomorrow.
[ 05 ]
The Prompt
One sharp, copy-pasteable prompt or workflow you can steal today. Small, practical, and tested before it ships.
From the Archive

Recent issues.

Browse all 248 issues →
No. 247 May 26, 2025

The eval problem nobody wants to talk about

Benchmarks are quietly broken, and most teams are shipping on numbers that don't mean what they think. Here's how to build evals you can actually trust.

Read issue → 5 MIN
No. 246 May 23, 2025

Open weights just changed the math for startups

When a capable model is free to run, the build-versus-buy decision flips overnight. We walk through the new economics with real numbers.

Read issue → 6 MIN
No. 245 May 22, 2025

Agents are getting good. Here's what breaks first.

Long-running agents fail in predictable ways. We map the five most common failure modes — and the cheapest guardrail for each one.

Read issue → 5 MIN
Why Inference

A newsletter that
respects your time.

Genuinely 5 minutes
We obsess over what to cut. If it doesn't change a decision you'll make this week, it doesn't make the issue.
No hype, no doom
No "this changes everything," no "the end is near." Just a clear-eyed read on what's real and what's noise.
Written by a builder
Every issue is written by someone who ships AI products — not a content team chasing a publishing quota.
Free, forever
No paywall, no upsell, no "premium tier." One honest sponsor a week keeps the whole thing free for you.
0K+
Subscribers
0%
Avg. Open Rate
0/5
Reader Rating
0
Issues Sent
Reader Mail

What readers say.

"It's the only newsletter I open every single day. Inference replaced about six tabs and two Slack channels for me — and somehow I'm more informed than when I was drowning in all of them."

RC
Rina Castellano
Founder, Halftone AI

"Most AI newsletters are either breathless hype or doom. Inference is the only one that reads like it's written by someone who actually ships code. The Tool Drop section alone has paid for itself ten times over."

DO
Daniel Osei
Staff ML Engineer, Ramp

"I forward it to my whole product team every morning. It's become our shared vocabulary for what's happening in AI. Five minutes, genuinely — and I never feel behind anymore."

MV
Maya Verma
Head of Product, Driftwave
DW
Dana Whitfield
WRITER & EDITOR · INFERENCE
The Writer

One person.
Every morning.

Before Inference, Dana spent eight years as a machine learning engineer — building recommendation systems, shipping models that didn't always work, and reading far too many papers at 2am.

Inference started in 2022 as a Slack thread for a handful of colleagues who kept asking the same question: "wait, what actually happened in AI this week?" It's now read by 182,000 people, and it's still written by one person, by hand, every weekday before 7am.

"My goal is simple: you read for five minutes, and you're the most informed person in your next meeting."
Tomorrow Morning · 7:00 AM

Be the smartest
person in the room.

Join 182,000+ builders who start their day with Inference. Free forever. Your first issue lands tomorrow.

No spam. No paywall. Unsubscribe in one click.