How to keep up with AI news without drowning in it
A one-hour-a-week system to keep up with AI, the sources worth reading for your role (checked in October 2026), and why job postings care.
Thomas Couderq, Founder of Tokn
8 min read
In this article⌄
- Why keeping up with AI matters if you want a job in AI
- Some postings ask for it in plain words
- How to keep up with AI: three sources and one routine
- Pick three sources, not thirty
- A one-hour weekly routine
- Primary sources first, hype last
- The best AI news sources by role (checked in October 2026)
- For engineers and AI engineers
- For researchers and research engineers
- For product, business and non-technical roles
- European picks
- AI podcasts worth your commute
- Turn the news into something you can show
- Where to go next on Tokn
- FAQ
- How much time should I spend keeping up with AI?
- What is the best AI newsletter?
- How do I keep up with AI research papers without reading arXiv every day?
- Do employers care whether I keep up with AI?
To keep up with AI, follow three sources, not thirty: one weekly analysis, one quick skim, and one primary feed (papers or release notes) for your field. Give it about an hour a week, plus 30 minutes trying something hands-on. That is enough to talk about recent tools in an interview and to spot what employers start asking for.
Staying up to date with AI is mostly a volume problem. On Tuesday 27 October 2026, arXiv's machine learning list received 226 new papers in a single day, plus 200 cross-listed from other fields. Nobody needs to read that. Below: why it matters for hiring, a weekly routine, and the sources worth your time by role, all checked as active in late October 2026. If you are earlier in the process, start with how to get a job in AI, step by step.
In short
- About 1 in 5 AI job postings on Tokn mentions agentic work, in both Europe and North America, and dozens name MCP, a protocol announced in November 2024.
- Pick three sources: one weekly analysis, one fast skim, one primary source. Drop the rest.
- One hour of reading a week, plus 30 minutes trying a new tool, beats a daily feed you never finish.
Why keeping up with AI matters if you want a job in AI
Because postings already name tools that are barely two years old, and some ask outright for people who follow the field. Here is what Tokn's postings showed on 27 October 2026.
| Term in the posting | Europe (793 postings) | North America (1,569 postings) |
|---|---|---|
| "agentic" | 173 (about 1 in 5), at 44 companies | 342 (about 1 in 5), at 38 companies |
| "evals" | 41, at 21 companies | 96, at 18 companies |
| MCP (Model Context Protocol) | 45, at 31 companies | 78, at 24 companies |
| An AI coding tool (Claude Code, Cursor, Codex, Copilot, Windsurf) | 27, at 17 companies | 41, at 18 companies |
Method: we fetched the full text of 2,362 of our 2,416 active postings from the companies' public ATS boards on 27 October 2026 and searched it for each term, excluding a tool maker's own postings, customer lists and company boilerplate repeated across a board.
Look at the dates. Anthropic announced MCP on 25 November 2024; less than two years later, 31 European companies name it. Claude Code became generally available on 22 May 2025, and OpenAI launched its Codex agent on 16 May 2025. Coding tool counts are small but spread across 17 to 18 companies per region. A Celonis AI deployment engineer posting asked for "comfort using AI-assisted tools in your work (e.g., Claude Code, Codex, Cursor, etc.), and the willingness to keep learning as they evolve." To choose one, read Claude Code vs Cursor vs Codex; for the wider picture, see the tools AI job postings ask for and the most in-demand AI skills.
Some postings ask for it in plain words
On the same date, 33 European postings at 18 companies and 26 North American postings at 12 companies explicitly asked candidates to keep up with AI research or tools, mostly in engineering and research roles. Examples advertised in October 2026:
- Cohere (data roles, North America): "you follow the research, have opinions, and enjoy being in the weeds".
- Harvey (agents engineering, North America): "Stay current on new developments in agentic systems and bring those learnings back to the products we build."
- Lovable (corporate communications, Europe), a non-technical role: "You keep up with what happens in the world instinctively — the news, the regulatory moves, the debates as they unfold."
How to keep up with AI: three sources and one routine
Few sources, a fixed weekly slot, and primary sources before commentary. The goal is not to know everything, but to know what matters for your work and to have tried it.
Pick three sources, not thirty
Choose one of each from the tables below.
- A weekly analysis of what mattered and why.
- A fast skim of what shipped.
- A primary source for your field: release notes if you build, papers if you research.
Even a newsletter publisher's own guide tells most readers to subscribe to two or three. We agree, and we would unsubscribe from anything skipped three weeks running.
A one-hour weekly routine
- Monday to Friday, 5 to 10 minutes: skim your fast source; open one link at most.
- One 20 to 30 minute slot: read your weekly analysis in full.
- Friday, 20 minutes: read one primary source properly, such as a paper, a model card or a changelog.
- Weekend, 30 minutes: try one new tool on a real task. Most people skip this step; it is the one that pays off in interviews.
Primary sources first, hype last
Read the release notes, paper or model card before the thread about it. Launch posts are claims: wait a week for independent evaluations before repeating a benchmark number. Don't read arXiv raw: Hugging Face Papers curates a daily list with community upvotes, and its Weekly tab is enough for most people.
Ignore "50 AI tools" lists, engagement threads and benchmark screenshots without a paper. X and LinkedIn are optional: follow a short list of builders and researchers, not the feed.
The best AI news sources by role (checked in October 2026)
These are the AI news sources we would pick, each checked on 27 October 2026 through its RSS feed or archive page. Frequencies are what we saw, not what publishers promise. Pick from your role's table only.
For engineers and AI engineers
| Source | Why read it | Frequency | Last item (27 Oct check) |
|---|---|---|---|
| Simon Willison's Weblog | A developer's hands-on tests of new models and tools | Near-daily | 27 Oct |
| Latent Space | AI engineering essays, news briefings and a podcast | Several posts a week | 27 Oct |
| TLDR AI | Five-minute daily skim; says it has 1,100,000 readers | Every weekday | 27 Oct |
| Claude Platform release notes and OpenAI API changelog | What changed in the APIs you build on | When features ship | 23 Oct (Claude) |
| The Pragmatic Engineer | How engineering teams really use AI | Weekly or more | 27 Oct |
Hands-on beats reading here: getting started with Claude Code is a good first weekend slot.
For researchers and research engineers
| Source | Why read it | Frequency | Last item (27 Oct check) |
|---|---|---|---|
| Hugging Face Papers (Weekly tab) | Community-filtered papers instead of raw arXiv | Daily list, weekly view | 27 Oct |
| Interconnects (Nathan Lambert) | Post-training and open models | About weekly, plus a podcast | 25 Oct |
| Ahead of AI (Sebastian Raschka) | Long technical explainers | Roughly monthly | 24 Oct |
| Import AI (Jack Clark) | Analysis of cutting-edge research | Weekly, Mondays | 26 Oct |
For product, business and non-technical roles
| Source | Why read it | Frequency | Last item (27 Oct check) |
|---|---|---|---|
| The Batch (DeepLearning.AI) | A calm summary of what mattered | Weekly | 23 Oct |
| One Useful Thing (Ethan Mollick) | How AI changes everyday work | Irregular, 2 to 3 a month | 20 Oct |
| Understanding AI (Timothy B. Lee) | How the technology works, clearly explained | About weekly | 19 Oct |
| Lenny's Newsletter ("How I AI") | How product people use AI in their jobs | Weekly or more | 26 Oct |
European picks
Europe has its own regulation and its own labs, and European employers expect you to know both.
| Source | Why read it | Frequency | Last item (27 Oct check) |
|---|---|---|---|
| Exponential View (Azeem Azhar) | Strategy and economics of AI | Several posts a week | 27 Oct |
| The EU AI Act Newsletter | The AI Act in practice, for anyone with European customers | Fortnightly | 20 Oct |
| State of AI Report 2026 (Air Street) | The big picture once a year (its monthly newsletter has been irregular since May) | Annual | Published 8 Oct |
| Mistral AI news and the Hugging Face blog | Releases from two of Europe's best-known AI companies | Several a month (Mistral), several a week (Hugging Face) | 21 Oct, 27 Oct |
AI podcasts worth your commute
Pick one: Latent Space if you build, No Priors or Dwarkesh if you care about strategy. Pair it with a written source, as audio is good for context and bad for detail.
| Podcast | Why listen | Frequency | Last episode (27 Oct check) |
|---|---|---|---|
| Latent Space | Builders and AI engineers talking shop | Two episodes in the two days before our check | 27 Oct |
| No Priors | Founders and investors on where AI products are heading | Two episodes in the week before our check | 23 Oct |
| The Cognitive Revolution (Nathan Labenz) | Long interviews with AI builders | Two episodes in the week before our check | 24 Oct |
| Practical AI | Applied AI for working engineers | Weekly | 22 Oct |
| Dwarkesh Podcast | Long, deep interviews, not all about AI | Every few weeks | 22 Oct |
Turn the news into something you can show
Reading helps your job search only once it turns into answers and projects.
- Keep a running note of what you tried this quarter. It feeds your interview answers.
- Build one small project per quarter with a new tool, and add it to a small AI portfolio.
- Before an interview, prepare two or three recent developments relevant to the company's product. See what ML interviews actually ask.
- Add the engineering blog of each company you apply to into your weekly slot.
Where to go next on Tokn
Browse agentic AI jobs to see how postings describe this work today, then set up a job alert so new openings join your weekly routine.
FAQ
How much time should I spend keeping up with AI?
For most people, 30 to 60 minutes of reading a week plus about 30 minutes hands-on. Researchers need more, mostly for papers. A fixed weekly slot beats an open-ended daily feed.
What is the best AI newsletter?
There isn't one for everybody. Pick one weekly analysis (Import AI or The Batch, say) and at most one fast skim (TLDR AI or Simon Willison's blog). Check that a newsletter still publishes before subscribing: some well-known ones have slowed down.
How do I keep up with AI research papers without reading arXiv every day?
Use a filter: the Weekly tab of Hugging Face Papers, Interconnects and Ahead of AI surface the papers worth your time. Then read one or two a week in full rather than skimming fifty abstracts.
Do employers care whether I keep up with AI?
Some say so explicitly. On Tokn, as of 27 October 2026, 33 European and 26 North American postings asked candidates to keep up with AI research or tools, and about 1 in 5 postings in both regions mentioned agentic work. We measure postings, not interviews, but discussing tools you have actually tried is easy preparation.
