Insights · Field notes
Our insights.
Field notes, benchmarks, points of view on AI in production. Written by the team, for teams that build.
[ 16 articles ]
All articles
15- Analysis
AI voice agents for business: what Mistral Voxtral 3 changes
Mistral's Voxtral 3 brings speech models and an SDK to build an AI voice agent for business. Per-minute pricing, EU hosting, limits: what SMEs should know.
8 min read/Anthony Cohen
- Analysis
AI agents and productivity at work: lessons from Anthropic
At Anthropic, AI agents now log six times more hours than staff. What the report says about AI agents and productivity at work, and what smaller firms can use.
8 min read/Anthony Cohen
- Analysis
Mistral Large 4 for Business: What Europe's New Model Changes
Mistral Large 4, unveiled on October 6, 2026, promises open weights by month end. What it means for business: cost, data sovereignty, hosting and its limits.
9 min read/Anthony Cohen
- Technical guide
n8n workflows + AI agents: the combo replacing brittle RPA projects
n8n handles the deterministic steps, the AI agent handles what's ambiguous and needs judgment. For most B2B automations that used to end up as brittle RPA projects, this duo genuinely changes the equation.
11 min read/Anthony Cohen
- Point of view
B2B Voice Agents: Why 2026 Is the Tipping Point
The voice agent technology stack shifted in 18 months. Accurate STT, natural TTS, sub-second latency: for N1 support, lead qualification, and appointment booking, the conditions are finally in place.
11 min read/Anthony Cohen
- Analysis
The real cost of an AI agent in 2026: a line-by-line breakdown
A worked example: a 200-person manufacturing company, a contract analysis agent, 50 active users. We break down every cost line, from API inference to monitoring, and compare year 1 to year 2.
11 min read/Anthony Cohen
- Technical guide
Evaluating AI agents in production: eval sets, drift, and monitoring
The agent works in the demo. Six months into production, how do you know it still works? The answer is structured evaluation, not infrastructure logs.
11 min read/Anthony Cohen
- Buyer's guide
Sovereign cloud or on-premise: the decision matrix for hosting your LLMs
SecNumCloud, public cloud, on-premise GPU: three options to host your LLMs, three very different levels of protection. Here is how to tell real data sovereignty from a sovereignty facade, and a decision matrix to guide your choice.
10 min read/Anthony Cohen
- Point of view
Why Your AI POC Never Made It to Production
The POC works, the demo convinces, yet the project never reaches production. It's not a model problem. These are three organizational blockers that most teams failed to anticipate.
9 min read/Anthony Cohen
- Technical guide
RAG on Internal Documents: Why Most Projects Stall
RAG promises an assistant that knows your documents. In production, it misses the mark, users disengage, and the project stalls. The root cause is not the model. It is a data quality problem.
11 min read/Anthony Cohen
- Regulatory analysis
EU AI Act: What Actually Applies to Your AI Deployments in 2026
The Digital Omnibus pushes EU AI Act high-risk obligations back to December 2, 2027. What already applies in 2026 and how to prepare your AI deployments.
11 min read/Anthony Cohen
- Point of view
Multi-agents: when it's useful, when it's pure overhead
Everyone is orchestrating agent swarms in 2026. Research shows it often makes things worse. Concrete criteria and a decision tree to get the architecture right.
10 min read/Anthony Cohen
- Security analysis
AI agent security: the real attack surface in 2026
Prompt injection, tool poisoning, silent exfiltration, RAG poisoning. What can break an AI agent in production today, and the layered defense framework we apply on every engagement.
13 min read/Anthony Cohen
- Buyer's guide
Local LLMs in 2026: which open-source model to pick for your enterprise
Mistral, Llama, Qwen, DeepSeek, Gemma: the local LLM landscape is rich but tricky to navigate. Our pragmatic buyer's guide for enterprises planning an on-premise deployment.
10 min read/Anthony Cohen
- Point of view
Sovereign note-takers: why big enterprises want 100% local
Large enterprises refuse consumer SaaS note-takers that ship their sensitive meetings to public LLMs. We build custom, fully local alternatives.
7 min read/Anthony Cohen