Date: {{current_date_full_with_day}}

Hey {{first_name | AI Visionaries}},

In this 108th edition, we look at the business model pressure building under AI: subscription economics and possible token price cuts.

Of course, nobody can stop talking about how great Fable is and how it was suddenly pulled back by Anthropic citing pressure from the US govt.

As usual, please send in your feedback and suggestions. It helps make the newsletter sharper for you.

-Renjit

PS: If you want to unleash the power of AI agents for personal and business productivity, you can setup time speak to me here »

Set up time to discuss:

AI subscriptions looked simple on the surface: charge a fixed monthly fee and let heavy users explore.

SemiAnalysis tested that assumption by buying Anthropic and OpenAI subscription plans, then running long coding tasks until the weekly limits ran out. The result was a reminder that power users can consume far more value than the sticker price suggests.

A $200 plan was widely assumed to top out near about $2,000 of API-equivalent tokens, but the tested plans appeared much more generous. That makes subscription margins highly sensitive to average usage, especially if API gross margins are around 75%.

For founders, the key point is not that subscriptions are doomed. It is that pricing, usage caps, and feature access are becoming strategic weapons. Labs may avoid bluntly cutting limits because users notice fast, but they can still hold back the newest models or advanced features for API customers.

SO WHAT?

If your product depends on using frontier models, plan for shifting access rules, not just shifting prices.

Read the source on X:

Most founders know their product. Few know how to get it in front of the right people. In this hands-on session, Clay + HubSpot for Startups walk you through ICP definition, prospect list enrichment, and AI-personalized outreach. You launch your first sequence before the session ends. June 18. 11am ET / 4pm GMT.

Register Before June 18.

OpenAI weighed major token price cuts as competition with Anthropic intensified, according to the Wall Street Journal. The timing matters because enterprise customers have started pushing back on fast-rising AI bills, especially for agentic and coding-heavy work.

Sam Altman acknowledged that cost had become a huge issue, and the company was looking for ways to help users get more value for less spend. The move would also be a direct response to Anthropic, whose Claude Code growth helped it win large enterprise customers and surpass OpenAI's valuation for the first time. The catch is margin pressure.

Frontier models still require expensive compute, and lower token prices could deepen losses even as both companies prepare for public-market scrutiny.

For founders, this is good news and a warning.

AI costs may fall, but customers will compare providers more aggressively because the products can feel interchangeable.

SO WHAT?

Treat model pricing as a moving market, and build your AI margins with enough room for a price war.

Read the source:

A widely shared post on X highlighted how Andrej Karpathy used AI in daily work: describe the task in plain language, let the system produce a first pass, inspect the result, then steer it with another sentence.

The post's point was simple but useful for founders. The core skill is not writing magic prompts; it is briefing an AI coworker clearly and checking the output with judgment. That framing turns AI from a chat box into an operating habit.

The next step is giving that instruction a schedule and the right tools so work continues after the first request.

The surrounding discussion also surfaced practical habits:

SO WHAT?

The real AI-native advantage is not occasional prompting; it is turning clear instructions, tools, and review loops into repeatable operating systems.

Read the source on X:

Stop spending more time double-checking your books than growing your business. BELAY provides the financial clarity and "peace of mind" you need to lead with confidence. Get your time back today. Get the Free Guide

Greg Isenberg's conversation with Theo Tabah framed an AI-native organization as one where people manage agents, agents read and write to the company, and the company gets smarter over time. That is a cleaner definition than simply giving everyone a chatbot. The strongest idea was the split between human judgment and machine execution: people own strategy, taste, communication, and review, while agents handle the repeatable middle of the work.

Theo walked through concrete workflows, including agent skill chains, a proposal microsite generated from company context, and a live feature prototype that fed usability feedback into a next version. The context layer was the real engine.

Agents could search, retrieve, and update shared markdown files, so every new task had more company memory behind it.

For founders, this points to a practical playbook: give agents clear goals, the right tools, useful context, and a quality bar.

SO WHAT?

AI-native teams will win by managing systems of agents backed by deep context, not by asking isolated chat questions.

See the video on YouTube:

Tell him the campaign. Viktor pulls last quarter's performance from Meta and TikTok, scrapes competitor ads, drafts the brief, posts it for review. You edit, he ships the creative requests to your designer. Inside Slack.

Get Started for Free

Anthropic launched Claude Fable 5 and Claude Mythos 5, then suspended access a few days later while saying it was working to restore availability. The launch itself showed how frontier AI is becoming a balancing act between capability, safety, pricing, and trust.

Fable 5 was described as Anthropic's most capable generally available model, with strong performance across software engineering, knowledge work, vision, scientific research, and long autonomous tasks.

Mythos 5 used the same underlying model with fewer safeguards in some areas and was aimed first at cyber-defenders and infrastructure providers through Project Glasswing.

Anthropic said Fable 5 included conservative safeguards that could route some requests to Claude Opus 4.8, with false positives in less than 5% of sessions on average.

Pricing was set at $10 per million input tokens and $50 per million output tokens, less than half the price of Claude Mythos Preview.

I was very lucky to use Fable for a complicated piece of work which it polished off to high quality and left me very usable artifacts. Moved my work forward by a couple of weeks!

SO WHAT?

The frontier is moving fast, but access reliability and safety controls now matter as much as benchmark wins. Does this mean we need to have a backup plan to have a local model?

Read the source:

Claude Managed Agents added two practical features: scheduled runs and secure environment variables in vaults. Scheduled deployments let an agent run on a cron-style cadence, start a fresh session each time, and complete recurring work without teams building their own scheduler.

Anthropic pointed to use cases like nightly data syncs, weekly compliance scans, daily digests, reports, log monitoring, sales follow-ups, and meeting reminders.

The vault update gives agents a safer way to use authenticated CLIs and services. Teams can register an API key with a variable name and approved domains, while the agent's sandbox only sees a placeholder.

The real key is attached at the network boundary, which helps keep secrets away from the model while still letting tools make approved requests.

Examples include automated Notion file uploads, Browserbase browser skills, KERNEL database checks, and Sentry-style operational workflows.

SO WHAT?

Agent products are moving from demos toward production plumbing: schedules, credentials, tools, and controls.

Read the source: