Import AI

What is Import AI?

Import AI is a weekly AI research newsletter written by Jack Clark. It is not an agent builder, API platform, or workflow tool. It is a reading product for people who want to understand what frontier AI systems are starting to do, what new papers actually mean, and how those changes connect to policy, safety, labor, and geopolitics.

Clark brings unusual context to the publication. He is a co-founder and Head of Policy at Anthropic, previously worked in policy at OpenAI, and before that was a technology journalist at Bloomberg. That mix shows up in the writing. Import AI does not read like a press release and it does not read like a dry paper summary either. It sits in the middle, translating technical work for readers who need both the details and the consequences.

The newsletter has grown into one of the best-known AI publications on Substack, with reported subscriber counts in the 96,000 to 120,000 range. Our research found it is read by researchers, executives, and policymakers, with issues typically taking about 12 minutes to read. For AgentsIndex visitors, the important point is simple: Import AI helps you understand the agent world, even though it is not a tool you use to build agents directly.

Key Features

• Weekly research analysis: Import AI publishes on a weekly cadence, which matters because it gives readers enough depth without turning into a constant news firehose. Instead of summarizing every headline, it picks a smaller set of developments and explains why they matter.

• Written by Jack Clark: Clark's background at Anthropic, OpenAI, Bloomberg, and the Stanford AI Index gives the newsletter a point of view that is hard to replicate. Readers are not just getting links to papers, they are getting interpretation from someone who has worked inside major AI institutions and policy debates.

• Free and paid access on Substack: The newsletter offers a free tier and a paid tier at $10/month. That keeps the main publication accessible to a broad audience while giving regular readers a simple way to support it.

• Frontier capability tracking: Import AI often highlights concrete capability jumps, not vague claims about progress. In recent coverage, it discussed research suggesting frontier models moved from succeeding on tasks worth 3 to 4 hours of human work to tasks worth roughly a week of work over about a year, a useful frame for anyone tracking agent progress.

• Safety and governance coverage: Many AI newsletters focus on product launches or startup funding. Import AI spends more time on model risk, cyber capability, policy responses, and institutional questions, which is useful if your job includes compliance, strategy, or public-sector work.

• Research examples with numbers: The newsletter regularly points to specific evaluations and benchmarks. One example from our research was coverage of AI cybersecurity work where frontier models were described as achieving about 50% success on tasks that would take human experts 3 to 4 hours, a much more useful signal than broad statements about "strong performance."

• Mix of technical and strategic framing: Import AI is one of the few publications that can move from a paper on reasoning or cyber capability into discussion of labor markets, national competition, or democratic institutions in the same issue. For readers who need to connect model behavior to business or policy decisions, that range is the real product.

Use Cases

Import AI is most useful when someone needs to make sense of fast-moving AI research without reading every paper themselves. A policy team tracking AI governance can use it to follow how new capabilities intersect with regulation and security. Clark's reporting and commentary are especially relevant here because he has testified on AI policy and helped shape discussion inside major labs, so the newsletter often reflects the questions policymakers are already wrestling with.

Researchers and technical leaders use it as a filter. In our research, recent issues covered topics like ARTEMIS, an AI agent scaffold for penetration testing, Numina-Lean-Agent for math reasoning, and MirrorCode-style software reimplementation benchmarks. Those are not generic "AI is improving" stories. They are examples of where autonomous systems are pushing into real work, and they help teams decide what to investigate next.

Executives and strategists use Import AI differently. They are less likely to care about every benchmark detail and more likely to care about what those benchmarks imply. When the newsletter discusses findings that frontier systems are moving from one-hour tasks toward multi-day tasks, that changes how leaders think about automation timelines, staffing, and product bets. It gives them a more grounded picture than vendor marketing.

For the AI agent community specifically, Import AI helps answer a recurring question: what can current systems actually do, and what is still hype? If you are building agents, evaluating tools, or deciding whether to trust a vendor's capability claims, the newsletter gives you a better map of the territory.

Strengths and Weaknesses

Strengths:

• Import AI has real author credibility, and readers feel that. A lot of AI newsletters summarize news from the outside. Clark has worked inside OpenAI and Anthropic, and that shows in how he frames capability progress, safety concerns, and policy tradeoffs.

• It is better than most newsletters at connecting technical work to consequences. Coverage of cyber evaluations, labor-market task duration, and governance questions gives readers a more useful picture than newsletters that just list model launches or funding rounds.

• The writing usually respects the reader's time. At around 12 minutes per issue, it is substantial without becoming an academic paper. For busy professionals, that balance is part of why the newsletter has reached roughly 96,000-plus subscribers.

• It covers the parts of AI that many business newsletters ignore. Safety, regulation, strategic competition, and benchmark design are recurring themes here. If you compare it with more product-focused newsletters like Ben's Bites or practical AI productivity newsletters, Import AI is much better for readers who need depth over speed.

Weaknesses:

• It is not beginner-friendly in the way many general AI newsletters are. Readers new to AI may find the assumptions too advanced, especially when issues jump quickly into benchmark interpretation, policy disputes, or technical paper details.

• It is a one-author publication, and that shapes both the tone and the limits. The upside is consistency. The downside is that coverage reflects one person's judgment about what matters most, rather than a broader editorial team with multiple specialties.

• It is not interactive. If you want a community, live discussion, office hours, or hands-on tutorials, this is not that kind of product. Compared with communities or research forums, Import AI is more of a one-way briefing.

• It is also not a tool in the practical sense. Some AgentsIndex visitors will arrive looking for software they can deploy. Import AI will not help you build an agent directly, so its value depends on whether you need understanding more than execution.

Pricing

• Free: $0
• Paid: $10/month

The free tier is the main reason Import AI has become such a widely read publication. You can subscribe and get the core newsletter without a software budget, procurement process, or trial period. For many readers, especially researchers and independent builders, the free version is enough.

The paid tier is simple at $10 per month, but our research did not surface a detailed breakdown of premium benefits. That means most people should treat the paid plan as support for the publication first, and extra access second, unless the current Substack page spells out more. Compared with software subscriptions, this is inexpensive. Compared with other newsletters, it sits in the normal range.

There are no usage credits, token fees, or hidden compute costs here, which is refreshing if you are used to AI tooling bills. The real cost is time. If you read it every week, you are spending attention to stay informed, and for the right audience that is a good trade.

Alternatives

The Batch from DeepLearning.AI serves readers who want a broad weekly overview of AI developments with a practical, educational tone. Compared with Import AI, it is less focused on policy and strategic risk. Someone might choose The Batch if they want a friendlier summary of the field, while Import AI is better for readers who want sharper interpretation of frontier research and governance.

Ben's Bites is built for speed. It is useful if you want startup news, product launches, and a quick sense of what people are talking about right now. Readers who care more about tools than research may prefer it. Readers who want to understand what a paper or benchmark means over the next two years will usually get more from Import AI.

TLDR AI works well as a daily scan. It helps people keep up with volume. The tradeoff is depth. Import AI is the better choice if you would rather read one thoughtful weekly issue than skim short summaries every day.

One Useful Thing by Ethan Mollick is often the better pick for educators, managers, and general knowledge workers trying to understand how AI changes day-to-day work. It is more accessible and more grounded in practical adoption. Import AI is stronger when the questions turn toward frontier capability, safety, and policy.

Direct paper reading and arXiv tracking is the real alternative for highly technical readers. If you already read papers every week, you may not need a curator. But many people still use Import AI because it helps them decide which papers matter and how to interpret them beyond the abstract and benchmark tables.

FAQ

What is Import AI?

Import AI is a weekly newsletter about AI research, policy, safety, and emerging capabilities, written by Jack Clark on Substack.

Is Import AI an AI agent tool?

No. It does not help you build or deploy agents directly. It helps you understand the research and strategic context around them.

Who writes Import AI?

Jack Clark writes it. He is a co-founder and Head of Policy at Anthropic, and previously worked at OpenAI and Bloomberg.

Who reads Import AI?

Our research found it is read by researchers, executives, policymakers, and technically curious professionals. Reported subscriber counts range from about 96,000 to 120,000.

How much does Import AI cost?

There is a free subscription tier and a paid tier at $10 per month.

Is the free version enough?

For many readers, yes. The free tier gives access to the core newsletter, which is the main reason people subscribe.

How do I get started?

Go to the Import AI Substack page and subscribe with your email. You can start with the free plan and decide later if you want to pay.

How long does it take to set up?

About a minute. It is just an email subscription, not a software installation.

How long does each issue take to read?

Roughly 12 minutes, based on our research. Some issues may take longer if you follow the linked papers and references.

Is Import AI good for beginners?

Only if you are comfortable learning as you go. It is readable, but it assumes some familiarity with AI concepts and current debates.

What makes Import AI different from other AI newsletters?

It combines paper analysis, policy thinking, and safety concerns in a way few newsletters do. Most alternatives lean harder toward product news, tutorials, or general business commentary.

Should AgentsIndex visitors care about it?

Yes, if they want to understand where agent capabilities are heading and how serious researchers and policymakers are interpreting those changes. No, if they only want software they can use today.