This month, we saw rapid developments in AI. including code agents shaking up cybersecurity stocks, free Google tools challenging photo startups, and some job titles disappearing from the radar.
Here’s a summary of the top stories, covering benchmarks, pricing, and their implications.
Figma and Anthropic turn AI code into editable designs
Figma teamed up with Anthropic on “Code to Canvas,” a tool that turns AI-generated code from Claude Sonnet 4.6 into fully editable Figma files. Developers ask Claude to create production-ready React or HTML/CSS user interfaces.
Figma then transforms this code into vector layers, components, and prototypes, eliminating the need for screenshot handoffs and manual recreations. Early users report up to five times faster iteration on AI-built applications.
The tool automatically detects interactive elements such as hovers, loading states, and modals, and supports multi-screen flows. Benchmarks show it beats earlier Claude models, with 70% of users preferring it over Sonnet 4.5 for contextual edits and 59% over Opus 4.5 for cutting unnecessary complexity.
It’s currently available to Figma Dev Mode users, making the platform a key part of AI-driven design workflows.
Google disrupts startups with Photoshoot AI
Google Labs launched Pomelli Photoshoot, a free tool that turns a single product photo into dozens of professional-quality images with custom backgrounds, lighting, and angles. Powered by Imagen 4, it recreates reflections, shadows, and fabric detailssimilar to studio shoots costing $500 to $5,000.
Currently available in the US, Canada, Australia, and New Zealand, beta integrations with Shopify and Etsy enable direct store updates. E-commerce users simply describe scenes, such as “neon-lit gym for sneakers,” to generate variants in seconds.
This launch challenges paid services like Booth AI and Pebblely, which charge $10 to $50 per image, and could take up to 80% of the $2 billion AI photo generation market.
Claude Code triggers $15B security stock sell-off
Anthropic previewed Claude Code Security, an AI that scans, patches, and verifies code for vulnerabilities on its own. A demo showed it fixing zero-day issues in live repositories faster than human teams, hitting 95% accuracy on the OWASP Top 10, including tough cases like memory corruption in curl and bash.
The news triggered a $15 billion drop in cybersecurity firms’ market value, with CrowdStrike down 11%, Palo Alto Networks 12%, and Zscaler 18%.
Capable of analysing entire GitHub monorepos, mapping data flows, and proposing tested pull requests, it integrates seamlessly with CI/CD pipelines. Companies like Atlassian and Vercel, among early testers, reported 70% reductions in production vulnerabilities.
Priced at $20 per user per month via Anthropic’s API, it undercuts traditional $100,000-a-year security roles, though it still occasionally falters against novel exploits.
Google launches AI music generator in Gemini
Google launched the Lyria 3 beta inside its Gemini apps, letting users create 30-second music tracks from text, image, or video prompts like “upbeat synthwave for a cyberpunk chase.” It supports track extensions, different vocal and style options, such as lo-fi or orchestral, and exports individual stems for digital audio workstations.
The feature is free for all Gemini users, with Pro offering longer tracks. It’s now available on web and mobile, including remixes for YouTube Shorts.
Lyria 3 scored 92% on human-likeness in blind tests with no melody hallucinations, backed by SynthID watermarks. Strong safety filters block harmful lyrics, though some worry about the training data used by YouTube.
Google announces Gemini 3.1 Pro
Google’s Gemini 3.1 Pro model has almost doubled its reasoning skills, scoring 77.1% on ARC-AGI-2 (up from 38%), 94.3% on GPQA Diamond, and 80.6% on SWE-Bench. It can handle million-token contexts for tasks like International Math Olympiad problems, kernel debugging, and physics simulations, beating OpenAI’s o1 in native tool use.
The API costs $2 per million input tokens and $12 per million output tokens. It’s available now on Vertex AI, with the consumer Gemini Advanced launching next week.
Meta patents AI “Digital Afterlife”
Meta filed patent #12,513,102 in December 2025 (inventor: Andrew Bosworth) for an AI system that simulates the behaviour of deceased or absent users based on Facebook and Instagram data.
It creates replies, stories, and video calls as an “eternal companion,” amid rising interest in grief technologies like Replika. Critics, including scholar Edina Harbinja, worry about consent and possible manipulation.
Meta says it’s for legacy preservation, with potential premium tiers or Horizon Worlds avatars.
European Parliament blocks AI tools on official work devices
The European Parliament has banned generative AI tools, including offline versions, on government-issued devices due to data leak risks, with fines of up to €35 million.
The policy affects over 10,000 staff. It allows email and documents but blocks writing assistants, summarizers, and web tools on tablets. It points to incidents like leaks from the German ministry and to audited local models like Llama.cpp, backing projects like Mistral.
“Software engineer” title could disappear in 2026
Boris Cherny, creator of Claude Code at Anthropic, forecasted that the traditional “software engineer” title could effectively vanish by the end of 2026, as AI agents automate 90% of routine coding tasks, from writing React components to debugging monorepos.
Speaking in recent interviews and X posts, Cherny argued that tools like Figma’s Code-to-Canvas integration (covered earlier this week) demonstrate production-ready handoffs from AI prompts, eliminating the need for junior-to-mid-level coding roles that dominated the past decade
