Building something that needs search data? SerpApi converts Google, Maps, Shopping, and other search engines into structured APIs. Just a simple GET request.

No scraping infrastructure.No proxy rotation.No parser maintenance.

Use cases include:👉 AI grounding👉 Market research👉 SEO, GEO, & rank tracking

Special offer for TLDR: Start with 250 free searches/month or claim 3 months at 50% off for new accounts.

Meta's new foundational AI model, Avocado, has fallen short on internal tests for reason, coding, and writing. It is not performing as strongly as the leading AI models from rivals like Google, OpenAI, and Anthropic. The model's release will be delayed until at least May while Meta works on improving it. Meta's leaders are already thinking about the next AI model, which will be named Watermelon.

Google Maps has implemented a feature called Ask Maps that helps users plan trips and answers complex questions about locations. Powered by Gemini, it is accessible via a button up near the search bar. It works like a chatbot, so users can send follow-up prompts to refine and expand on its suggestions. Ask Maps is rolling out now in the US and India, but only on Android and iOS. Web support will come later.

Data-first home robot company Sunday Robotics has announced a Series B raise. The company's unique approach claims zero robot data, with a focus on real-world data only. It uses exact duplicates of tools mounted on robot arms to collect all of its robot data while data collectors wear a hat that captures wide field-of-view RGB camera images. The company has developed a glove that allows it to create a distributed network of operators to collect high-quality, diverse data at a massive scale.

Factory-manufactured homes can, in certain cases, yield substantial cost savings over conventional construction. A single-wide manufactured home is probably around 40% to 50% cheaper than an equivalent site-built home. However, these savings quickly disappear with larger and more complex homes. This doesn't mean the industrialization of home building is a doomed enterprise, but it does show how it is an enormously complex problem.

You've built the perfect fine-tuned model - but running it on generic cloud infrastructure, you're going to hit performance bottlenecks. The culprit? Standard infra that wasn't designed for proprietary architectures.

Crusoe's Bring Your Own Model (BYOM) is your ticket to scale without compromise. Powered by MemoryAlloy™ technology, Crusoe optimizes its inference engine specifically for your unique weights, achieving:

→ 5x higher throughput.

→ Breakthrough Time-To-First-Token speed compared to stock vLLM.

→ Dedicated capacity and "glass box" visibility.

→ Total IP ownership stays with model builders.

Stop letting infrastructure hold back your AI innovation. Scale your fine-tuned models with precision

The Perplexity API platform replaces model providers, search layers, and embeddings. It allows developers to swap between the latest frontier models and configure curated presets, tool access, step limits, and token budgets, all from one endpoint. Developers have direct access to the same index that powers Perplexity to call real-time information with citations within apps and agents. They will also have access to the same execution environment that Perplexity uses internally as a standalone service.

Shopify's Liquid is an open-source Ruby template engine somewhat inspired by Django. Its developer found dozens of new performance micro-optimizations using a variant of Andrej Karpathy's Autoresearch, a system that runs hundreds of semi-autonomous experiments to find new effective training techniques. Implementing the optimizations resulted in a 53% improvement on benchmarks, an impressive improvement for a code base that's been tweaked with hundreds of contributors over 20 years. The experiment shows how effective the Autoresearch pattern can be.

A huge amount of modern work is already hollow. People sit in meetings that don't matter, produce things that won't last, and perform a version of productivity that everyone silently pretends is real. Most mass layoffs aren't really due to some AI revolution, but because those jobs were already obsolete. AI is bringing in an era where the constraints of work that told us what to do, who to be, and how to spend our days are dissolving. It will take us some time to learn how to deal with this freedom.

The market instability in February was a preview of what it feels like when the increasing ability of AI starts to interact with markets, jobs, and governments all at once. While feelings of uncertainty will likely spread further, uncertainty is not the same as helplessness. The choices that individuals and organizations make now matter because there is still time to influence the direction the technology develops. Every organization trying to figure out a good way to use AI right now is setting a precedent for everyone else.

He will remain as Adobe's company chair.

This post describes a vibe coding system that uses types as specifications, which forces compilers to reject code until they can produce a valid proof that the output satisfies the specification.

As the first engineer on TLDR's new Applied AI team, you'll build AI agents and composable Claude Skills to let non-technical teammates create their own AI workflows. Learn more.

Vite 8 delivers up to 10-30x faster builds while maintaining full plugin compatibility.

Xbox mode is a controller-optimized interface that lets players browse their library, launch games, use Game Bar, and switch between apps.

Amazon's internal AI tool, Kiro, frequently hallucinates and generates flawed code that software developers have to spend time fixing.

There is still plenty of engineering work in software engineering, even if it looks different.

Most dictation tools create more work. You speak, then spend five minutes fixing what went wrong.

You'll see the difference instantly with Wispr Flow: clean, sendable text as fast as you can speak. It's a life changer:

• 4x faster than typing. Emails, Slack threads, docs, code comments. All dictated in real time.
• Works in every app. System-wide on Mac, Windows, iOS, and Android. Zero setup, zero context switching.
• Millions of users worldwide. Not a side project. The voice tool people actually switch to.

No cleanup. No copy/paste. Just talk and send.

Try Wispr Flow Free →

Apple's new MacBook Neo is the first Mac to be powered by an iPhone chip. Using the A18 Pro that debuted in 2024's iPhone 16 Pro models, the MacBook Neo is 50% faster for everyday tasks than the best-selling PC with the latest shipping Intel Core Ultra 5. Available in Silver, Indigo, Blush, and Citrus color options, the device has a 13-inch Liquid Retina display without a notch. It has a 16-hour battery life, 8GB of unified memory, two USB-C ports, and a headphone jack. The MacBook Neo starts at $599 for 256GB of storage and is available for preorder now, with availability beginning on March 11.

Google will lower its Play Store fee to 20%, and 15% for new installs of apps from developers participating in its new App Experience program or updated Google Play Games Level Up program. Subscription fees have been lowered to 10%. Google will now charge developers in the UK, US, and European Economic Area a 5% fee for using its billing system and a market-specific rate in other regions. Developers can offer alternative billing systems or guide users to their own websites for purchases. Third-party app stores can now apply to Google's Registered App Stores program, which gives them a streamlined installation interface in Android if they meet certain quality and safety benchmarks.

The US Nuclear Regulatory Commission has issued a construction approval to TerraPower to begin working on a site in Kemmerer, Wyoming. TerraPower, financially backed by Bill Gates, is attempting to build a sodium-cooled reactor that incorporates energy storage as part of its design. The project is expected to be completed in 2030, but delays are expected as it is a first-of-its-kind project. It is being built as a joint public-private partnership as part of the Department of Energy's Advanced Reactor Demonstration Program.

Evo 2 is an open-source AI model trained on genomes from all three domains of life. It features internal representations of key features in complex genomes like regular story DNA and splice sites, which can be challenging for humans to spot. The model is fully open, including model parameters, training code, inference code, and the training dataset. It is useful for evaluating genomes and identifying key features.

Through the end of Q1, get a free consultation with an experienced DuploCloud Solution Architect. They'll evaluate your AWS architecture, processes, and governance, showing you where you can optimize, secure, and save.

This review normally costs $2,500. Sign up before the end of March to get it free.

The Google Workspace CLI is a CLI for all of Google Workspace designed for humans and AI agents. The tool builds its entire command surface dynamically at runtime using Google's Discovery Service, so new API endpoints and methods are picked up automatically. It supports multiple auth workflows, so it works on PC, in CI, and on a server. There are currently over 100 Agent Skills for APIs, common workflows, and Google apps.

Every new generation of agents will force engineers to rethink what is optimal. In this situation, less is more. Using too many libraries and harnesses locks engineers into a solution for a problem that may not exist in future generations of agents. If a real problem existed, and there was a good solution for it, frontier companies would be the biggest users of that solution, and they would incorporate that solution into their product. It's possible to do ground-breaking work now with just basic CLI and understanding a few basic principles about agentic engineering.

The runaway intelligence case probably won't happen because the amount of electricity we can generate and the number of chips we can make is pretty limited. Models will certainly be useful in creating the next generation of models in the same way computers are useful tools when designing the next generation of computers, and it seems reasonable to expect that we'll eventually make an AI that is better than any human at anything. However, it is very possible that artificial intelligence will either plateau, just like human intelligence, or we'll run out of resources to compute the next iteration of intelligence.

Crypto firm Kraken's banking unit has won access to the Federal Reserve's core payment systems. The approval of the master account will allow Kraken Financial to handle transactions more quickly and seamlessly for big clients and professional traders. The approval is a major win for the crypto industry, which has been rebuffed for years in its efforts to tap into the system. It makes Kraken the first crypto firm to be able to move money on the same rails used by thousands of banks and credit unions.

Over 2M+ clients have made meaningful progress with a Grow provider, whose mission is to bridge the gap between "having coverage" and actually getting mental health support. Apply to work at Grow.

Giving new package versions about a week on a registry before installing them can block most supply chain attacks, as it gives the community and security vendors time to flag problems.

Agents and developers are now able to request Markdown output from WordPress.org.

CLIs are the lowest-friction interface for AI agents to reach external systems.

The value of writing code is falling, but the value of knowing what to build is rising.

The things that survive in a world where software is cheap to produce are positions, earned trust, and permissions that took years to acquire.

If 'AI-rewriting' is accepted as a valid way to change licenses, then any developer could take any GPL-licensed project, get an LLM to rewrite it in a different style, and release it under MIT.

Building something that needs search data? SerpApi converts Google, Maps, Shopping, and other search engines into structured APIs. Just a simple GET request.

No scraping infrastructure.No proxy rotation.No parser maintenance.

Use cases include:👉 AI grounding👉 Market research👉 SEO, GEO, & rank tracking

Special offer for TLDR: Start with 250 free searches/month or claim 3 months at 50% off for new accounts.

OpenAI's GPT-5.3 Instant is a conversational model designed to have smoother conversations. The model has improved contextual understanding of user queries, so it can answer more questions without assuming bad intent without any loosening of safety controls. This also makes it better at web searches. The model generates more useful direct responses rather than presenting users with lists of links.

Apple has added the M5 Pro and M5 Max to its MacBook Pro lineup. The M5 Pro now comes with 1TB of storage by default, while the M5 Max comes with 2TB of storage by default. The base storage for the M5 MacBook Pro has increased to 1TB. The upgrades come with increases to starting prices. Apple has also introduced a pair of new displays.

The Jesse-Victor-Gharabaghi (JVG) quantum decryption algorithm completely upends existing time projections for quantum decryption. It requires a thousand-fold less quantum computer resources to break the encryption methods used in RSA and ECC than Shor's approach. Shor's algorithm requires an estimated one million qubits, but JVG requires less than 5,000. Researchers previously believed we were a decade away from quantum decryption, but that timeline has now moved up significantly.

Astrolab and Interlune are forming a partnership to build a fleet of electric harvesters to excavate, extract, and separate Helium-3 from lunar regolith on the Moon. The companies have not announced when they are scheduled to deploy an initial harvester, but both are working toward that goal. Astrolab's FLEX rover will likely be one of the first payloads on the first SpaceX Starship mission to the lunar surface. Helium-3 doesn't occur naturally on Earth. It has several applications, but the most near-term use is in cryogenics.

Humans already couldn't keep up with code reviews when humans wrote code at human speed, so there's no way we can manually review AI-generated code at scale, especially when reviewing AI-generated code requires more effort. Developers need to move upstream and focus on encoding the constraints. The future is ship fast, observe everybody, and revert faster.

Engineering teams praise engineers who overbuild, but the ones who ship simple things get nothing. This is what happens when companies evaluate work incorrectly. Complexity looks smart, and our systems are set up to reward it. Engineers need to learn how to describe the judgment behind their work to show how decisions were made and why complexity was avoided.

AI skeptics often argue that while AI can pretend to be people, it shouldn't. However, the biggest AI labs build human-like systems because that is the best way to build a capable AI system. Personalities are the medium that make language models useful. They guide models to pull from the helpful parts of their training data.

Apple's Pro and Max chips usually take the same basic building blocks from the basic chip and just scale them up. However, the M5 Pro and Max chips use a different CPU architecture. The new Fusion Architecture welds two silicon chiplets into a single processor. Rather than just a pair of M5 chips welded together, Apple has one chiplet handle the CPU and most of the I/O, while the other chiplet is used mainly for graphics. Apple is also introducing a new type of 'performance core' that prioritizes multi-threaded performance.

Native no longer provides performance and visual benefits compared to web.

OpenAI decided to create its own GitHub after its engineers encountered a rise in service disruptions.

The humanoid robot can autonomously perform tasks such as loading self-tapping nuts at assembly stations and transporting material boxes.

Chrome will move from a four-week release schedule to a two-week release schedule starting in September.

The next generation of Starlink Mobile will deliver broadband capabilities to unmodified cell phones.

It's a smarter choice to be an individual contributor in 2026 than an engineering manager.

If QA is a bottleneck on your software engineering team and you're releasing slowly because of it, you must check out QA Wolf.

Their AI-native service gets engineering teams to 80% automated end-to-end test coverage, and helps dev teams ship 5x faster by reducing QA cycles from hours to minutes.

QA Wolf takes testing off your plate. They can get you:

• Unlimited parallel test runs for web and mobile
• 24-hour maintenance and on-demand test creation
• Human-verified bug reports sent directly to your team
• Zero flakes guarantee

The benefit? No more manual E2E testing. No more slow QA cycles. No more bugs reaching production.

With QA Wolf, Drata's team of 80+ engineers achieved 4x more test cases and 86% faster QA cycles.

Schedule a demo to learn more

Apple has introduced a new iPhone in March for the second year in a row. The iPhone 17e is a basic, no-frills iPhone that features an A19 chip with four GPU cores and an Apple C1X cellular modem. The phone supports Apple Intelligence and MagSafe charging. It will be available for preorder on March 4 for a March 11 launch. The base 256GB model will cost $599, while an upgraded 512GB model will be available for $799.

Meta is testing a shopping research feature in its AI chatbot. The feature responds with carousels of product images with captions containing information about brands, websites, and prices. The chatbot's recommendations are tailored to users' location and other details when applicable. It is unclear whether Meta currently receives referral commissions for its chatbot recommendations or whether the AI prioritizes brands that already advertise with Meta.

SpaceX will attempt to catch Starship's upper stage with its launch tower after two perfect soft landings in the ocean. Starship V3 Ship 1 is now headed for ground tests. The vehicle is designed for full reusability. The first Starship V3 flight of 2026 will likely occur this month.

A research team at UC Davis Health has safely performed a spina bifida treatment combining fetal surgery with stem cells. The therapy involved adding a layer of human placenta-derived stem cells to standard fetal surgery. Putting stem cells into a growing fetus was a total unknown, but the team reported great safety. The success paves the way for new treatment options for children with birth defects.

In the open-weight ecosystem, teams build on each other's innovations to compound the pace of progress. This post looks at various open source models and the engineering bets that define each one. Every major open-weight model released at the frontier since 2025 uses a Mixture-of-Experts transformer architecture. Almost every 'open source' model is actually open weight - this means the trained parameters are public, but the training data and often the full training code are not.

A harness is a layer that connects, protects, and orchestrates components without doing the work itself. Every agent framework is trying to build its own harness from scratch, but durable, event-driven infrastructure already solves this. Utah (Universally Triggered Agent Harness) is a conversational Telegram or Slack agent with tools, memory, sub-agent delegation, and full durability - a durable, cloud-ready OpenClaw. It was created to prove that durable, event-driven infrastructure can be used to create an agent harness that decouples orchestration from the agentic loop.

US policy has become increasingly unpredictable, thuggish, arbitrary, and capricious. Even if threats against Anthropic are scaled back, great damage has been done. The government has shown that it will treat homegrown companies like a foreign adversary simply for refusing to capitulate to its terms of business - for having its own opinions on how to, or how not to, deploy its own property. Companies cooperating with the US government will now have to operate under this logic.

Intercom found success as a late-stage SaaS company in AI. Three years ago, its future was looking pretty bleak as the company was heading quickly towards negative growth. It launched a new service agent called Fin in the summer of 2023, after which its business began to violently recover. The process involved destroying many parts of the business and creating new things. It was difficult, but the company is now at $400M ARR overall, with overall growth rates doubling for the past two years.

Voice-to-text that works in every app. Millions of users worldwide. Free on Android during launch. Try free

SpaceX's upcoming Starlink V2 satellites will deliver up to 100 times the data density of the current first-generation system.

The strikes caused structural damage, disrupted power delivery to infrastructure, and in some cases, required fire suppression activities that resulted in additional water damage.

Stripe is introducing a new feature that will allow companies to charge an automatic markup on token usage.

The subscriptions that will survive are the ones that can't exist in any other way.

Go exists in a sweet spot of usability, performance, and ubiquity that makes it a very good fit for agents.

The US government was always going to use AI for whatever purposes it desired, regardless of provider - the way to address this is with new laws and through strengthening accountable oversight.

Building something that needs search data?

SerpApi converts Google, Maps, Shopping, and other search engines into structured APIs — returning reliable JSON you can plug directly into your app.

No scraping infrastructure.No proxy rotation.No parser maintenance.Just a simple GET request.

Use cases include:👉 Market research👉 SEO, GEO, & rank tracking👉 AI grounding

Special offer for TLDR: Start with 250 free searches/month or claim 3 months at 50% off for new accounts.

SpaceX may file for its initial public offering as soon as this month. The firm expects to submit its draft IPO registration to the US SEC in the next few weeks, keeping it on track for a June listing. It could be seeking a valuation of more than $1.75 trillion. The listing could raise as much as $50 billion.

Emil Michael, the Department of War's chief technology officer, had been negotiating with Anthropic for weeks on a $200 million AI contract. The talks had hit an obstacle as Anthropic would not allow its technology to be used for purposes such as the surveillance of Americans. Michael demanded that Anthropic's CEO, Dario Amodei, get on the phone to discuss the issue, but he was told Amodei was in a meeting with his executive team and needed more time. Michael had been hammering out a deal with OpenAI on the side, so the Department of War designated Anthropic as a security risk and cut it off from working with the government.

Google has implemented a new Merkle Tree Certificate system in Chrome. It uses Merkle Trees to provide quantum-resistant assurances that a certificate has been published without having to add most of the lengthy keys and hashes. Cloudflare is enrolling roughly 1,000 TLS certificates to test how well the system works. The Internet Engineering Task Force standards body has formed a new group to develop a long-term solution for generating the distributed ledger.

Google is building a new data center in Pine Island, Minnesota, that will be powered by 1.9 gigawatts of clean energy from wind and solar. It will use a 300-megawatt iron-air battery with a 30-gigawatt-hour capacity and 100-hour duration. Iron-air batteries store electricity by making rust and then reversing the rusting process to release energy when needed. While these batteries are heavier and less efficient than their counterparts, they are almost three times cheaper. Google will fund the installation of many smaller batteries across the grid to increase power capacity and make the system more reliable.

DevOps is one thing for a small team and another entirely for enterprises. In this eBook, Datadog shares 9 dev enablement practices that have been proven to scale DevOps without sacrificing excessive resources. Learn how to create reporting standards and provide tools that elevate software quality. Get the eBook

Designing tools for models is as much of an art as it is a science. It depends heavily on the model, the goal of the agent, and the environment it's operating in. Developers need to experience often, read outputs, and try new things. This post provides a useful framework for designing agents with tools that are shaped to their own abilities.

Apple plans to replace Core ML with a modernized Core AI framework to help developers better leverage modern AI capabilities with their apps in iOS 27. The shift from 'ML' to 'AI' is an acknowledgement from Apple that 'machine learning' is a dated term that no longer resonates with developers or consumers. The general purpose of Core AI will remain the same, but there will likely be new ways to incorporate AI features into apps. Apple will announce iOS 27 at its WWDC conference in June.

Anthropic's Claude has dethroned OpenAI's ChatGPT in Apple's App Store. The sudden surge in rankings is almost certainly due to public backlash at a recent decision by OpenAI to work with the Department of War to deploy AI without restrictions on mass domestic surveillance and fully autonomous weapons. Anthropic was recently labeled a supply-chain risk to national security for refusing to amend its safety policies for its work with the government.

OpenAI CEO Sam Altman recently held an Ask Me Anything on X to discuss the company's recent decision to work with the Department of War. In the thread, he answered questions around whether society should prefer a democratically elected government or unelected private companies to have more power, whether the government should nationalize OpenAI or other AI efforts, and the work it takes to keep society safe. Katrina Mulligan, OpenAI's Head of National Security Partnerships, and Boaz Barak, a computer scientist at OpenAI, were also present in the thread to answer questions.

Cloudflare's Agents platform allows developers to build agents in an environment designed for durable execution, serverless inference, and pricing that scales.

This post contains notes on building APIs in an AI-first world from someone who has maintained many APIs but only has basic working-professional knowledge of building APIs at scale, securing APIs, and related subjects.

Humanoid robots are like self-driving cars, but significantly more complex, and we still haven't fully solved self-driving, so it is unlikely that we're close to having general-purpose humanoid robots at home.

Cursor is moving towards helping developers build the factory that creates their software.

It's important to distinguish between what's scientifically true and what's empirically useful.

Amazon's winning formula for many years has been to sell the thing everyone wants but charge less for it.

Flow turns your voice into clean, sendable text inside any app. Speak naturally and Flow handles the cleanup. 89% of messages sent with zero edits. 4x faster than typing.

Millions of people use Flow daily, including teams at OpenAI, Vercel, and Clay. Works on Mac, Windows, iPhone, and now Android.

Free and unlimited on Android during launch.

Download on Android | Get Flow Free

Nano Banana 2, also known as Gemini 3.1 Flash Image, delivers results similar to Nano Banana Pro but with the speed of the non-pro Flash variant. The new image generator has more advanced world knowledge, which gives it the necessary information to render objects with greater fidelity and create more accurate infographics. Nano Banana 2 will replace both the standard and Pro variants of Nano Banana across the Gemini app, Search, AI Studio, Vertex AI, and Flow. Example images generated with the model are available in the article.

Netflix is walking away from a deal to buy Warner Bros. Discovery's (WBD) studio and streaming assets as its board deemed on Thursday that a revised bid by Paramount is a superior offer. Paramount raised its bid earlier this week to $31 per share in cash. Netflix had granted WBD a seven-day waiver to re-engage with Paramount last week. Paramount's offer is for the entirety of the company.

OpenAI's VP of Science, Kevin Weil, recently had an interview where he discussed how AI capabilities are progressing to the future of robotic labs, how founders should think about building with AI, and why this might be the best moment in history to start a company. This post contains highlights of the interview along with a link to a video of the chat. Weil is now focused on accelerating scientific discovery with AI. His previous experience includes leading product teams at Twitter, Facebook, and Instagram.

Doctors may soon prescribe pills that can travel through the digestive system and check for tissue health. Diagnosis for gastrointestinal diseases is often complicated, and treatment can bring serious side effects. Capsules are able to handle much of that work and could even release drugs exactly where they're needed or snip biopsy samples for analysis. The technology could transform care and make screening for disease less invasive and more accessible.

As more developers let Claude Code handle tool selection, the stacks that the AI chooses become 'the stack'. The model's training data may shape market share more than a marketing budget or conference talk. This study looks at what tools Claude Code actually picks. Claude Code would rather build from scratch than use existing tools - custom implementations are its single most common recommendation.

The dynamics of the software industry are rapidly changing. Some of the foundations that open source was built on are looking increasingly shaky. Open source maintainers have been fighting companies taking their work and creating competing products for years by using increasingly restrictive licenses or keeping certain parts of the development stack private. However, AI is making it easier to clone codebases into entirely new products. Software licenses may become meaningless as AI becomes adept enough to rewrite open source codebases from scratch without using any of the original code.

Block plans to lay off 40% of its workforce - more than 4,000 employees. Founder Jack Dorsey alluded that AI tools were the reason for the cuts in a letter to shareholders. He claims that the decision to cut jobs wasn't because the company is in trouble. Block has delivered mixed results in recent years as it faces tough competition.

Anthropic's CEO, Dario Amodei, says that the company cannot, in good conscience, agree to allow the Department of Defense to use its AI models in all lawful use cases without limitations. The Department of Defense wants to be able to use Anthropic's models without restrictions on fully autonomous weapons and mass domestic surveillance. Anthropic has refused to change its stance despite threats by Defense Secretary Pete Hegseth to label it as a 'supply chain risk' or invoke the Defense Production Act to force it to comply with the department's demands. Discussions between the two parties are still ongoing.

Voice-to-text that works in every app. Millions of users worldwide. Free on Android during launch. Try free

Leverage agents make users orders of magnitude more productive, while function agents perform tasks for users.

The conditions that the economy must experience for AI to produce a sustained negative demand shock are unlikely.

The world's greatest living mathematician Terence Tao's approval of generative AI seemingly legitimizes the great promise of the technology.

Many things that look like exponential functions, like the adoption of new technologies, actually follow an S-curve.

Open source models and similarity among products may ultimately keep AI from financial success.

Victor Nechita, the Vehicle Program Manager for the Cybercab, is leaving Tesla after six years at a critical moment for the program.