AI & Models9 min read
Grok Bot Is What Demand for Intelligence Looks Like
Who’s talking about SpaceXAI’s new agent, who’s actually using it, and what the token bill looks like once the thing stays on overnight.
AI & Models
Frontier labs, model releases, capability trends and the economics of inference.
28 articles
AI & Models8 min readWhy AI Needs Open Models
Open-source AI gives startups, developers and businesses an alternative to relying entirely on a small group of closed model providers. Restricting access to open models could reduce competition, increase costs and limit how companies build and deploy AI, while doing little to stop the technology from advancing elsewhere. That matters even more as the model itself becomes just one component of a larger system that increasingly determines how useful an AI product actually is.
AI & Models4 min readKimi K3
DeepSeek showed that frontier-level intelligence could be built and served far more efficiently than the market expected. Kimi K3 takes that disruption further by combining competitive performance with open weights, giving developers and enterprises more control over how models are hosted, customized and deployed. The right response is not to restrict access, but to build stronger open alternatives and compete in a model market that is becoming cheaper, more open and far less defensible.
AI & Models7 min readAI 2040
"AI 2040: Plan A" proposes a radical blueprint to prevent a reckless global race to superintelligence through full research transparency and enforceable U.S.–China cooperation. By pausing AI capability growth at human-expert levels until 2040 using "mutually assured compute destruction," the plan ai
AI & Models4 min readChina May Be Ending the Open-Model Era
For years, China’s leading AI labs used open-weight models to catch up with their American rivals. Now that some of those models are approaching the frontier, Beijing may be reconsidering whether the rest of the world should have access at all. The next phase of the AI race may be shaped not just by
AI & Models11 min readA Better Way to Decide Which AI Model to Use
The best AI model is no longer simply the smartest one. As providers flood the market with frontier, mini, flash, reasoning, coding, long-context, closed, and open-weight models, the real question is how much useful intelligence each model delivers relative to cost, latency, compute, context, and op
AI & Models3 min readIntelligence Per Dollar
For years, the AI race has been defined by one question: who has the smartest model? But as open source models like Zhipu's GLM 5.2 approach frontier performance at a fraction of the cost, the metric that matters is shifting from raw intelligence to intelligence per dollar. The next stage of AI comp
AI & Models8 min readThe Future of Agent Orchestration: Skill-Based Control Planes
AI architecture is fundamentally shifting from monolithic, hard-coded agents to modular, skill-based control planes that intelligently orchestrate specialized capabilities. Using InferX and the Model Context Protocol (MCP) as a case study, this new paradigm enables dynamic provider abstraction, plug
AI & Models5 min readWho Connects the Machines?
AI adoption is no longer limited by model intelligence alone. As agents become more capable, the next bottleneck is connectivity, creating demand for shared protocols like MCP that allow AI systems to interact seamlessly with tools, data, and workflows. Just as HTTP and TCP/IP helped transform the i
AI & Models5 min readThe Harness, Not the Model
What matters is the agent harness, the terminal that unlocks value of the underlying model. As businesses, and individuals continue to increase compute consumption, the harness becomes of the highest value ways to create higher quality tokens instead of fud.
AI & Models2 min readWhy Your Best Employees Should Be Expensive
AI changes the old SaaS logic that more usage always equals better value, because with token-based pricing higher employee usage can raise costs while also driving much greater output. The argument is that top performers should be allowed, and even encouraged, to spend heavily on AI tools when that
VE Weekly
The week in tech, AI and early-stage capital.
One send a week. Whole pieces, not a digest of links out.