Analysis
217 articles
Analysis and opinion on where AI is going and what it means for a business that has to make decisions now.
Page 2
Showing 21 to 40 of 217
The alpha in served intelligence: what 155 trillion tokens a day really tells you
DeepSeek open sourced DeepInfer and disclosed one cluster serving 155 trillion tokens a day at US$38 of revenue per GPU-hour. We checked the maths against its public list prices and it closes. The margin is in the KV cache, not the model, and that changes where the opportunity sits for Australian teams.
Is Complex Agentic Memory Always Needed? Why Naive RAG Beats Agentic Memory Systems
A raw-history RAG baseline beat 12 agentic memory systems in six of seven benchmark settings, at a fraction of the ingestion cost. Here is what the evidence shows and when a memory layer still earns its place.
Kimi K3.1 Pro vs Claude Fable 5.1 and GPT-6 Astra: Open Weights Have Caught the Frontier on Paper
Kimi K3.1 Pro scores 55 on the Artificial Analysis Intelligence Index, two points above Claude Fable 5.1 and GPT-6 Astra, at $1.98 per task against $7.63 and $3.26. Here is what the open-weights milestone proves, and what the benchmark audit says it does not.
Do No-Code AI Agent Platforms Beat a Custom Built Agent?
No-code agent platforms are fast and cheap to start, then credits scale with volume. We compare both paths on cost, integration depth and governance.
What Happens When an AI Agent Job Runs for Hours in the Background?
MCP Tasks replaced the long-held connection with a durable task handle. Here is the lifecycle, the support gap, and how to design a job that survives a crash.
Should You Let AI Agents Run Your HR and Finance Workflows?
Workday and Google now put agents inside HR and payroll. The limiting factor is not the model, it is the process map and the approval chain.
Can Local AI Agents Really Replace Your Cloud Chatbot Subscriptions?
We ran the numbers on $20 seats versus a $6,950 local agent box. Payback lands at 15 months for 25 seats, and the trade-offs are not only financial.
Who Is Your AI Agent When It Logs Into Your Bank Account?
US banks are blocking third-party AI agents that borrow customer credentials. Here is the identity, scope and audit model that makes agent access safe.
Google Agent Compute Engine: What It Is and Why Every AI Agent Gets Its Own microVM
Google Agent Compute Engine gives every AI agent a hardware-isolated microVM with zero credentials inside. How it works, what it costs, and how it compares with Agent Substrate and AWS Bedrock AgentCore.
LocalRouter on One DGX Spark: A Full-Duplex Local AI Assistant, Measured
JJ Eaton's LocalRouter runs chat, speech, image and video on a single DGX Spark. The project's own numbers show a 961 millisecond voice loop, a 14 MiB scheduler, and licences that stop it shipping commercially.
MiniMax M3.1 Flash vs DeepSeek V4.1 Flash 1017 vs GLM-5.3-Flash: which Flash model is cheapest per finished task?
MiniMax M3.1 Flash, DeepSeek V4.1 Flash 1017 and GLM-5.3-Flash all land within four Intelligence Index points, but cost per finished task ranges from US$0.22 to US$0.40. A sourced three-way comparison of benchmarks, licence terms and DGX Spark self-hosting.
Should A Cloud Agent Run Your Business, Or Should You Keep It On-Prem?
Claude Duet puts a proactive agent in Anthropic's US cloud. Here is how Australian small businesses should split work between a cloud agent and an on-prem agent, with the real costs and the privacy rule that applies.
Recursive Flint and the Eureka Machine: How AI Built a 40B Coding Model in Four Months
Recursive Flint is a 40B open-weight coding model that decodes at 150 tokens per second on a desktop. We break down the benchmarks, the Eureka Machine research loop that built it, and what it means for local coding agents.
tsrs: The Rust Port That Made TypeScript 8x Faster and Got Microsoft to Ship It
An agent-written Rust port of TypeScript is 11.81x faster at the same checker count, and Microsoft is already shipping its optimisations. Here is what the numbers mean.
DeepSeek V4.1 Flash on dual DGX Spark: the best local setup for an AI agent
We run Hermes Agent on DeepSeek V4.1 Flash across two DGX Sparks. Measured on our own endpoint: 35.1 tokens per second on a single stream, 61.7 tokens per second at four concurrent requests, 1,713 tokens per second of prefill, and 111 MB of KV cache for a 131k context. Here is why this pairing is the best local agent setup we have run, and where it still loses to the cloud.
SKY-001: What Skyfall's Engineering World Model Proves, and What It Doesn't
Skyfall's SKY-001 claims 51x faster and 150x cheaper CAD than GPT-6 Astra. We separate the verified numbers from the vendor claims.
Google just open-sourced self-improving agents. The harness beat the model.
Google Cloud AI Research open-sourced Harness², a framework that lets an agent rewrite its own harness around a frozen model. GLM-5 with a one-step self-improved harness beat Gemini 3.5 Flash with its stock one, 75.7 vs 72.4 on WorkBuddy-Bench.
If a Jailbroken PS5 Can Run an LLM, What Is Your Excuse?
A jailbroken PS5 runs llama.cpp at 55 tokens per second on a 9B model, outperforming an RTX 4060 on bandwidth. Hardware is ready for local AI. Vendor API gatekeeping is the real blocker.
NIST's 2026 AI Agent Warning: What SMBs Are Missing (And How to Fix It)
NIST's February 2026 AI Agent Standards Initiative mandates three governance pillars: agent identity, runtime monitoring, and supply chain integrity. Most SMBs lack all three.
If AI Agents Work 6x More Hours, Why Do Most Automations Still Fail?
Anthropic proved 6x agent hours. Most organizations capture 10-30% of that gain in year one. Here is the capture-rate formula that tells you what your agents are worth.
What's actually useful in AI, every Friday
We follow everything happening in AI and send what's worth your time: new models, AI agents, and how to put them to work. One email a week, unsubscribe anytime.