AI + Careers Briefing — October 22, 2026

Today's stories share one thread: AI is being put to work on real tasks, not just conversations. It helped researchers find and verify a pair of critical web flaws, it is being invited (or blocked) as a shopper at the biggest U.S. retailers, it is speeding up how a carmaker designs vehicles, and one startup says it largely built a new coding model. For job seekers, the common question is the same: where do people still direct, check and finish the work?

What is PHP2Shell, and why are governments warning about it?

PHP2Shell is the name for two critical vulnerabilities in PHP, the widely used open-source web programming language, that could let hackers remotely execute malicious code on affected servers, according to CNBC.

The U.S. National Security Agency urged organizations using PHP on Wednesday to prepare security updates, warning the flaws could give attackers direct access to server file systems and databases. The FBI issued a similar warning, and authorities in Singapore, the United Kingdom, Australia, Canada and the European Union issued alerts. Amazon said it detected attempts to exploit the flaws before public disclosure, but it is unclear how attackers learned of them. Google said its Wiz unit used AI models to help find and investigate the flaws, and its Gemini AI helped assess them, test potential attacks and verify patches. OpenAI said researchers using its models also helped uncover vulnerabilities, and that it spent two months leading a coalition to coordinate fixes. The PHP team has released security fixes, and not every PHP site is necessarily vulnerable.

Why it matters for job seekers: The episode shows AI becoming part of how security researchers find and verify flaws, while people still coordinate fixes and apply patches. If you are targeting security, DevOps or web development roles, we would expect patching and vulnerability-management experience to be worth highlighting, along with comfort using AI tools in that work. Our guide to AI-proof career skills covers how to frame these skills.

Source: CNBC

Why are Walmart and Amazon split on AI shopping agents?

Walmart is opening its site to AI shopping agents through a new Agent Gateway, while Amazon has begun blocking agents from making purchases for online shoppers, according to TechCrunch.

Walmart announced the Gateway on Wednesday, and CEO John Furner discussed the plans at the Masters of Scale Summit in San Francisco. It is an MCP server, built on the model context protocol that Anthropic originally created, and it lets approved agents search Walmart's online inventory and buy on a user's behalf. It supports shopping at both Walmart and Sam's Club. Walmart says Agent Profiles, rolling out next month, will let customers authorize each agent separately, view its purchases apart from their own, and either approve every transaction or set a spending limit. Walmart is also introducing terms for agents. Its returns process applies even when an agent does not act as intended, agents cannot access payment information, and agents that break the terms will be blocked. TechCrunch reports Walmart already announced a partnership with Meta's Muse and offers integrations with ChatGPT, Gemini and other AI apps. It adds that various travel and booking sites have locked agents out entirely.

Why it matters for job seekers: The split suggests retailers are making different bets on agent-driven shopping, and that could shape what e-commerce, product and marketing teams hire for. We would expect roles touching checkout, customer controls, returns policy and partner integrations to come up in interviews. Reviewing this announcement, we would be ready to explain how you would design or measure a shopping experience when an AI agent, not a person, is the buyer. Our guide to landing a remote product management job covers how to talk about product work like this.

Source: TechCrunch

How is GM using compute and AI to compete with Chinese automakers?

General Motors is rebuilding its next vehicles around a single central computer and AI-assisted engineering, aiming to halve its development timeline and match Chinese rivals, TechCrunch reported after the company's GM Forward event on Oct. 21.

GM's fourth-generation EV platform centers on new battery chemistries, revamped electric motors and a new central computing platform and electrical architecture. TechCrunch reports the central compute unit will be 40x more powerful than GM's existing systems, and that removing smaller chips should trim wiring by 30%. The overhaul is set to debut in 2028 with the Cadillac Escalade IQ, alongside an autonomous highway driving feature that GM says will allow hands-off, eyes-off driving. TechCrunch reports GM's capitalized software amortization expense was $866 million in 2025.

Chinese automakers have shortened development cycles to two to three years, down from the usual four to five, according to TechCrunch. GM says enterprise AI tools have sped up vehicle development and wants to cut its traditional four- to five-year timeline in half. In one example, engineers went through 300,000 virtual designs of a new lithium-manganese-rich battery pack before assembling the first physical prototype, and the concept moved to an approved program in one month.

Why it matters for job seekers: The article points to automotive hiring leaning toward software, compute and simulation work alongside traditional vehicle engineering, though it does not report any GM hiring plans. If you target this sector, we would expect employers to value experience with software-defined systems, AI-assisted design and fast iteration. Our guide to skills-based hiring covers how to show that kind of range to employers.

Source: TechCrunch

What is Recursive's Flint coding model?

Flint is a free coding model from AI startup Recursive that runs on a laptop and, the company says, was built largely with the help of AI.

Recursive released Flint on Oct. 21, according to TechCrunch. The company says its automated research system, the Eureka Machine, helped with nearly every step of development, which let a small team build the model from scratch in under four months. Flint is a mixture-of-experts model that uses 2.5 billion of its 40 billion parameters at a time. It is free to download on Hugging Face under the Apache 2.0 license. Because it runs on a user's own computer, developers can use it offline, keep their code private and avoid cloud usage fees, TechCrunch reports.

In Recursive's tests, Flint generated about 131 tokens per second on a MacBook Pro with an M4 Pro chip, roughly twice the 58 to 69 that similar models managed on the same machine. On SWE-Bench Verified, it outperformed other models under 50 billion parameters. Recursive acknowledged Flint needs more guidance than the big paid coding assistants and suggested breaking large tasks into smaller steps. Wired's Will Knight, who tried it on a Mac with 48GB of memory, found it handled his coding requests every time.

Why it matters for job seekers: These are Recursive's own test results, and they do not show how Flint performs on real team codebases. Still, free models that run offline could change which tools employers allow and how developers work. We would expect interviewers to ask how you use AI tools and how you check their output. Reviewing this launch, we would practice breaking large tasks into smaller steps and explaining what you verified yourself. Our guide to technical interview prep is a place to start.

Source: TechCrunch

What to watch

• Whether PHP2Shell exploit activity grows as organizations apply the PHP team's fixes.
• Walmart's Agent Profiles, which the company says roll out next month, and whether other retailers follow Walmart or Amazon.
• GM's fourth-generation platform ahead of its planned 2028 debut in the Cadillac Escalade IQ.