TLDR 2026-08-24

40% of companies heavily using AI say inflation actually helped their business this year… only 12% of non-AI adopters say the same. That contrast turned up everywhere in Mercury's survey of 1,500 early-stage founders & builders: the companies leveraging AI the most are raising more, hiring differently, and reporting a different macroeconomic picture.

Read Mercury's new report to explore:

• The 31-point confidence gap between heavy AI adopters and everyone else
• What startups spend on AI each month
• What junior hiring looks like at AI-forward companies and beyond

Plus more on how founders are spending, raising, and hiring in 2026. Read the report.

Apple is set to introduce its foldable iPhone at a launch event on or around September 9. Samsung has stolen some of the thunder from the launch by introducing a similar design, but that just reinforces that the new form factor is a winner. Apple's device reportedly excels as a camera viewfinder, but it lacks a telephoto camera, which could be an issue for early adopters spending over $2,000. The design also relies on Touch ID rather than Face ID.

Some of Nvidia's biggest customers have been told that the price of the AI servers containing its flagship Vera Rubin and Grace Blackwell chips is going up by more than 15% due to the soaring costs of memory chips. The price hikes will go into effect early next year. The increases will depend on the generation of Nvidia chips and the memory configurations. Nvidia has also raised prices for its gaming-oriented PC graphics cards.

Amazon's all-electric toaster-shaped Zoox taxis are the first robotaxis offering rides on the roads without standard driver features like a dashboard or pedals. They can drive in either direction and hold up to four passengers sitting facing each other. Amazon has started rolling out the vehicles across San Francisco and Las Vegas. Tesla's Cybercab lacks a steering wheel and pedals, but it has not received the same regulatory approval as Zoox to begin offering rides to the public.

The opening day of the World Humanoid Robot Games in Beijing on Saturday saw Chinese humanoid robots break records set by humans, including Usain Bolt's 100-meter sprint world record. More than 2,000 humanoid robots participated in the event. There were 51 events and more than 1,000 competitions, including running, table tennis, and soccer. While the games showcased humanoid robots' rapidly evolving abilities, experts say it will still take time to achieve mass real-world deployment and that these robots are mostly still used for demonstrations, performances, and research.

Kent C. Dodds built Kody, and he's taking it apart live with Serge from Sentry:

• the agentic loops powering kody.codes
• Cursor automations
• Sentry webhooks + Seer
• what's next for Kody

Bring questions. They're answering them live. Register free

A lot of the tedium it used to take to get a tricky optimization working is now gone. Many optimizations that were previously too risky to try make sense to try out now. AI drastically reduces the cost of optimization. It can turn days or weeks of work into a trivial loop.

The updated roadmap for the Model Context Protocol (MCP) sets the direction for protocol work over the coming months. It is organized into five priority areas: agentic messaging primitives, HTTP-native transport unification and hardening, agent identity and enterprise-ready security, improved primitives, and improved SDK developer experience. Specification Enhancement Proposals that fall within these priority areas get expedited review and have the best chance of acceptance.

New news is coming out so fast that it seems crazy for a founder not to be fully plugged in. Being online can help build a presence and keep people as informed as possible. Posting frequently can help drive inbound interest. Tech Twitter can provide a view of emerging consensus, but few immediately actionable insights. It's important to actively maintain cognitive security and cognitive hygiene, as spending too much time online can be dangerous.

TLDR is hiring a GTM Engineer to join our Applied AI team and own our AI-native GTM stack. We're looking for someone comfortable building AI agents and working with HubSpot. Click here to learn more!

When Moore's Law was in full effect, it didn't make sense to ruthlessly optimize code, as a new CPU would arrive in 18 months that would double performance. Developers only started to have to think about parallelization, architecture, memory locality, etc, after Moore's Law started slowing down in the mid-2000s. Prior to Fable, it felt like a waste to spend too much time improving coding harnesses or context strategies, as a new model would arrive at the same price or cheaper in just a few months that would cover most of the problem. Fable is incredible, but the cost is so high that developers are now starting to think about what work should go where and optimizing their harnesses to provide weaker models with sufficient context to perform well.

Buildkite runs agents on your own infra — any cloud, any OS, any chip, secrets never leaving your network. Uber and Shopify run 1.4B job-minutes a week. Try Buildkite free

Access models from 80+ providers with better prices, automatic routing and fallbacks, and no subscriptions. Build with the right model for every request. See how it works.

A good coding harness can help improve code quality dramatically.

The front end is less risky to just hand to an agent, so many AI coders will feel comfortable just letting their agents handle it unsupervised.

Finding problems comes from staying engaged with people's work long enough to see what no single request can show you.

Nvidia will invest $1 billion in AI startup Poolside at a pre-money valuation of $12 billion and will pay $6 billion to license Poolside's technology and hire the bulk of its engineers.

AI enables developers to pick less common languages and work with much harder technologies.

Table/game selection is just as important as execution, especially in markets.