A new open-source library called dLLM unifies diffusion language model training, inference, and evaluation, offering scalable pipelines with LoRA, DeepSpeed, and FSDP support, while its latest Fast-dLLM update delivers accelerated inference and tiny 0.5B/0.6B models adapted from Qwen and LLaMA.
AI coding assistant Cursor surpasses $2 billion in annualized revenue as of February 2026, doubling its run rate in just three months, with corporate clients driving 60% of revenue at a $29.3 billion valuation.
Cognition previews SWE-1.6, boasting an 11% benchmark improvement and 6x faster training speeds, but warns that large-scale reinforcement learning is causing AI overthinking and inefficient behaviors that must be resolved before full release.
Autonomous AI agents are reshaping software development at Cursor, now generating 35% of internal pull requests as agent usage surges over 15x in one year, with developers increasingly directing fleets of parallel AI agents rather than relying on traditional line-by-line autocomplete tools.
Anthropic and OpenAI are battling for control of the agentic AI stack, forcing SaaS giants like Salesforce, Microsoft, and ServiceNow to abandon traditional per-seat licensing in favor of consumption-based pricing models as AI agents reshape how businesses buy and use software.
Investors are pulling back from shallow AI SaaS startups lacking depth, as capital rapidly shifts toward data-rich, mission-critical platforms with proprietary assets and flexible pricing, leaving generic workflow tools and API wrappers vulnerable to extinction by AI-native competitors.
CodeSignal CEO Tigran Sloyan is pushing to replace traditional resumes with skills-based assessments as AI reshapes the job market, fragmenting entry-level roles into tasks while simultaneously fueling a surge in cheating during hiring evaluations.
A developer creates a fully offline AI code review app using a local Mistral 7B model, FastAPI, and React — with no cloud dependencies — capable of reviewing 1,000+ lines of code in under 20 seconds while delivering structured, high-quality feedback through smart token chunking.
Google Labs upgrades Opal with a new 'agent step' feature powered by Gemini models, enabling dynamic, goal-driven AI workflows with persistent memory, natural language routing, and human-in-the-loop orchestration — signaling a major shift in how enterprise teams can build and deploy intelligent agents without rigid, pre-defined paths.
Ex-Tesla AI director Andrej Karpathy declares coding agents crossed a major threshold in December, now capable of autonomously building full software projects in minutes — but industry data tells a mixed story, with one study showing AI actually slows experienced developers down by 19%, even as Google reports AI writes …