All stories
An Agent’s Memory Is Now a Deliberate System, Not a Longer Prompt
AI agents are moving beyond ever-longer prompts. This explainer shows how compaction, retrieval, deterministic tools, subagents and durable state create more reliable, efficient memory for long-running work.
Deepgram’s Daily Deal Tests Whether Open Voice AI Can Stay Neutral
Deepgram’s acquisition of Daily expands its voice AI infrastructure reach while testing whether Pipecat can remain genuinely open, vendor neutral and independently governed after the deal.
Mistral Puts a Trillion-Parameter Cyber Model in Customers’ Hands
Mistral has released the weights for its trillion-parameter Large 4 model, giving cybersecurity teams an open-weight alternative to commercial APIs while raising questions about cost, control and dual-use risk.
Wabi Bets $1.25 Billion on an AI Agent That Builds Your Next App
Wabi has raised $55 million at a $1.25 billion valuation to develop an AI agent that turns group conversations into personalized mini apps, while facing questions about trust, traction and privacy.
Cohere targets CAD$7 billion in funding at a CAD$35.7 billion valuation
Cohere is targeting up to CAD$7 billion in Series E funding at a CAD$35.7 billion valuation, while questions remain over the initial close, final terms and the AI company’s growth prospects.
GPT-6 Makes Reasoning Effort the New Unit of AI Economics
GPT-6 reframes AI economics around reasoning effort, model routing, token costs, prompt caching and adaptive budgets, helping companies balance quality, latency and operational expense.
Flint’s Real Test Is Whether AI Can Shrink the AI Development Race
Recursive’s Flint is a laptop-friendly coding model built with extensive AI assistance. Its benchmark results are promising, but independent reviews question whether speed translates into reliable everyday software development.
GitHub Copilot Fusion Bets That Model Routing Can Cut Coding Costs
GitHub Copilot Fusion routes tasks across Claude and GPT models, promising lower coding-agent costs while raising questions about billing, benchmark limits, workspace changes and human review.
Google’s Gemini 4 Carbon Lead Is Hard to Test and Harder to Buy
Google’s Gemini 4 Carbon leads a major AI benchmark, but restricted access, no public API and limited technical disclosure leave enterprises unable to test its real-world value.