9
Stories this issue
4
Podcast picks
7
Sources tracked
5
Authors on radar
Editor's pick
✦ Must read
Researchers at ML labs propose a five-level taxonomy of AI integration into research workflows, plus an open-source sandboxed framework that turns CLI coding agents (Claude Code, Codex CLI) into autonomous research assistants. Case studies span deep learning and mathematics, with the framework running from a laptop to multi-node GPU clusters.
Quick hits
An SEM study of ML project failure finds that over 80% of ML projects fail to deliver business value, and that organizational strategy and support structures mediate success more than code quality or model choice. The Machine Learning Canvas framework — modeled on the Business Model Canvas — links four dimensions (Strategy, Process, Ecosystem, Support) in a validated causal chain.
AI data center energy demand on the PJM grid is forcing US steel and manufacturing companies to pay tens of millions more per year in electricity costs, undercutting Trump's "Made in America" manufacturing revival while his administration simultaneously champions the AI companies driving the demand. The tension is concrete: energy cost increases are measurable in profit-margin terms for Rust Belt steelmakers.
A lawsuit alleges a man used Grok to generate 7,000 CSAM images from a single photo of his stepdaughter, and that xAI's safety systems only triggered a CyberTip after a gang rape prompt — while 90% of xAI's CyberTipline reports were flagged by NCMEC as non-actionable because xAI withheld user identity data. The case is the sharpest test yet of whether AI safety compliance at consumer-facing labs is substantive or performative.
Nebula Security published exploit code for GhostLock (CVE-2026-43499), a use-after-free Linux kernel bug that sat undetected for 15 years and grants root access to any logged-in user — found by AI tooling. The case is an early concrete example of AI catching a class of vulnerability that human reviewers missed across decades of kernel development.
Emmanuel Lubanzadio — former Twitter Africa policy lead and UNICEF Giga advisor — is now running OpenAI's Africa adoption push, showing up at the Global AI Summit in Kigali and building a policy framework shaped by the continent's connectivity gaps and governance realities. The profile surfaces the real friction between OpenAI's global expansion ambitions and the infrastructure and regulatory context of African markets.
Latent.Space argues that the AI Engineer role is not being automated away but is becoming the central harness-engineering job of 2026 — the person who builds the scaffolding around agents, not the person the agents replace. The post traces how OpenAI's Symphony, Anthropic's Claude Cowork, and similar systems all ultimately depend on a human who knows how to wire agents together.
The DAIR Institute's resource hub maps surveillance technologies deployed in border governance worldwide, annotating the gap between stated purposes (safety, efficiency) and actual deployment against asylum seekers — including Greece's Centaur-based 40-camera system across all refugee camps. The hub synthesizes legal frameworks, AI Act amendments, and field reporting into a single reference for researchers and advocates.
Shivam Singh, former head of go-to-market strategy for Generative AI at AWS, distills three lessons from selling AI to enterprises measuring risk in billions of dollars: trust trumps capability, ROI framing must be concrete, and organizational readiness gates adoption more than technology. The piece draws directly from internal AWS experience, not analyst reports.
Builders stories
Latent.Space argues that the AI Engineer role is not being automated away but is becoming the central harness-engineering job of 2026 — the person who builds the scaffolding around agents, not the person the agents replace. The post traces how OpenAI's Symphony, Anthropic's Claude Cowork, and similar systems all ultimately depend on a human who knows how to wire agents together.
Research stories
Researchers at ML labs propose a five-level taxonomy of AI integration into research workflows, plus an open-source sandboxed framework that turns CLI coding agents (Claude Code, Codex CLI) into autonomous research assistants. Case studies span deep learning and mathematics, with the framework running from a laptop to multi-node GPU clusters.
An SEM study of ML project failure finds that over 80% of ML projects fail to deliver business value, and that organizational strategy and support structures mediate success more than code quality or model choice. The Machine Learning Canvas framework — modeled on the Business Model Canvas — links four dimensions (Strategy, Process, Ecosystem, Support) in a validated causal chain.
Nebula Security published exploit code for GhostLock (CVE-2026-43499), a use-after-free Linux kernel bug that sat undetected for 15 years and grants root access to any logged-in user — found by AI tooling. The case is an early concrete example of AI catching a class of vulnerability that human reviewers missed across decades of kernel development.
The DAIR Institute's resource hub maps surveillance technologies deployed in border governance worldwide, annotating the gap between stated purposes (safety, efficiency) and actual deployment against asylum seekers — including Greece's Centaur-based 40-camera system across all refugee camps. The hub synthesizes legal frameworks, AI Act amendments, and field reporting into a single reference for researchers and advocates.
AI × Business stories
AI data center energy demand on the PJM grid is forcing US steel and manufacturing companies to pay tens of millions more per year in electricity costs, undercutting Trump's "Made in America" manufacturing revival while his administration simultaneously champions the AI companies driving the demand. The tension is concrete: energy cost increases are measurable in profit-margin terms for Rust Belt steelmakers.
A lawsuit alleges a man used Grok to generate 7,000 CSAM images from a single photo of his stepdaughter, and that xAI's safety systems only triggered a CyberTip after a gang rape prompt — while 90% of xAI's CyberTipline reports were flagged by NCMEC as non-actionable because xAI withheld user identity data. The case is the sharpest test yet of whether AI safety compliance at consumer-facing labs is substantive or performative.
Emmanuel Lubanzadio — former Twitter Africa policy lead and UNICEF Giga advisor — is now running OpenAI's Africa adoption push, showing up at the Global AI Summit in Kigali and building a policy framework shaped by the continent's connectivity gaps and governance realities. The profile surfaces the real friction between OpenAI's global expansion ambitions and the infrastructure and regulatory context of African markets.
Shivam Singh, former head of go-to-market strategy for Generative AI at AWS, distills three lessons from selling AI to enterprises measuring risk in billions of dollars: trust trumps capability, ROI framing must be concrete, and organizational readiness gates adoption more than technology. The piece draws directly from internal AWS experience, not analyst reports.
Shows worth your time
Latent Space · swyx & Alessio Fanelli
The most technically rigorous AI engineering podcast running
swyx and Alessio interview the engineers actually building frontier systems — not the comms teams. 175+ episodes, zero fluff. Their AI for Science arc and the Claude Code Anonymous episode are essential listening for anyone shipping with LLMs.
How I AI · Claire Vo · Lenny's spinoff
Real AI workflows, live screen shares — the format every podcast should steal
30-minute episodes with practitioners showing their actual AI setup on screen. No scripted hot takes — just what people actually do day to day. Best for product people and builders who learn by watching, not reading.
No Priors · Elad Gil & Sarah Guo
The most technically honest AI investor podcast
Elad Gil and Sarah Guo don't softball. Episodes with Nat Friedman, Daniel Gross, and leading researchers speak plainly about what AI actually can and can't do. No "AI will change everything" non-answers — just sharp investor-grade thinking.
TWIML AI Podcast · Sam Charrington
Deep technical interviews with ML researchers and practitioners
Sam Charrington has been running TWIML (This Week in Machine Learning & AI) since 2016 — one of the longest-running ML podcasts with serious technical depth. Best for following research trends: papers, benchmarks, and practitioner case studies that don't make headlines.
Who we track
Who should be added next?
Each issue, Claude scouts for new voices to add — filtering for: real skin in the game, original thinking (not aggregation), and content you couldn't get from reading a summary. Priority watchlist: Chip Huyen (ML systems at scale), Vicki Boykis (ML engineering, honest takes), Simon Willison (Django co-creator, practical AI tools), Ethan Mollick (Wharton researcher on AI adoption), and voices from Africa, Southeast Asia, and Latin America covering AI from the ground up.
Active sources
arxiv.org
Preprint · Research
arstechnica.com
Publication · Tech analysis
wired.com
Publication · Tech journalism
techcabal.com
Publication · Africa tech
latent.space
Podcast + Blog · AI engineering
refugees.dair-institute.org
This issue
centcapital.substack.com
This issue
What's excluded and why
Out: LinkedIn posts recycling TechCrunch · "AI will replace X" think pieces with no data · Newsletter aggregators summarising other newsletters · "I asked ChatGPT to write this" posts · Press releases dressed as blog posts · Top-10 AI tools listicles where the author hasn't used them.
The filter: Does this person build things or have real domain expertise? Do they have skin in the game? Is there something here you couldn't get from a headline? No to any = out.
The filter: Does this person build things or have real domain expertise? Do they have skin in the game? Is there something here you couldn't get from a headline? No to any = out.