Insights - What we learn building AI

Practical lessons from real AI projects. Engineering depth, business clarity.

AI and privacy in Switzerland: what business owners need to know

The nFADP already applies to your AI tools. A practical guide to what Swiss data protection law requires before you deploy.

Not everything needs to be an MCP tool

MCP is great for what it does. But we turned two tool calls into structured JSON and saved 40% on tokens.

The AI chatbot cost checklist we wish we had earlier

We cut our chatbot cost from $0.15 per request to near zero. Five checks that made the difference.

How we use Claude Code hooks to keep context clean

Stop dumping every rule into CLAUDE.md. Hooks inject tool-specific context only when the agent actually needs it.

What actually eats your context window

Most developers guess wrong about what costs tokens. A quiz format walkthrough of the biggest context window misconceptions.

Building an AI Research Platform: ETL, RAG, and a Chatbot That Actually Knows Your Data

How we built a research data platform that ingests data from APIs, CSVs, and public databases into a unified schema, then lets researchers chat with it using RAG and MCP.

How to Write Tools That AI Agents Can Actually Use

Most AI agent projects fail because of bad tool definitions, not bad models. Here is how to write API descriptions that agents understand and use correctly.

How We Built an Agentic Claims System: Deterministic First, AI Second

An AI system for car insurance claims that validates, enriches, and evaluates. The key insight: deterministic checks must run before the LLM ever sees the claim.

What Goes Wrong When AI Systems Work Together

Anthropic found that AI agents create identical mistakes, start turf wars, and trust bad data. Here is what that means for your multi-agent plans.

Making AI Chat Indistinguishable from Human: What We Learned

We researched what makes AI text detectable, tried fine-tuning on WhatsApp exports, and found that the real tells are not words but patterns. Here is everything we learned.

How to Evaluate AI Output (When You Can Not Check Every Answer)

LLM evaluation is the hardest part of AI in production. Here is how we measure quality at scale without manually reviewing every response.

Setting Up AI for Your Team: Developers vs Business Users Need Different Things

We set up AI coding tools for dev teams and AI-assisted development for business users. They require completely different approaches. Here is what works for each.

Context Engineering: Why Your AI Gets Better Answers Than Your Competitor's

Prompt engineering is last year. Context engineering is what separates AI systems that work from those that hallucinate. Here is the difference and why it matters.

RAG in Production: What Actually Works

We processed millions of documents with RAG pipelines. Here is what we learned about retrieval quality, chunking strategies, and the mistakes that cost us weeks.

Automating Job Applications at Scale: Architecture of an AI-Powered Pipeline

How we built a system that scrapes, scores, tailors, and submits job applications automatically. The architecture decisions, what worked, and what surprised us.

When NOT to Use AI: An Honest Guide

As an AI consultancy, we talk clients out of AI projects regularly. Here are the situations where AI is the wrong answer and what to do instead.

Scaling AI Agents: Why You Need to Separate Brain from Hands

When your AI agent grows beyond a demo, the architecture that got you there will break. Here is the pattern Anthropic uses and why it matters for your project.

The Real Cost of Running AI in Production

Most AI budgets are wrong by a factor of 3-5x. Here is what actually costs money, and how to cut 40-70% without losing quality.

AI for Swiss Financial Services: What Works Today

From compliance automation to client advisory, here is where AI delivers real value in Swiss banking and insurance, and where FINMA sets the boundaries.

AI and Your Workforce: What the Research Actually Says

Not hype, not fear. Here is what peer-reviewed research tells us about AI productivity gains, job transformation, and what successful retraining actually looks like.

Why Simple Beats Smart: Choosing the Right AI Architecture

Anthropic themselves say simple patterns beat complex frameworks. Here is how to pick the right level of AI complexity for your problem.

How to Build AI Agents That Run for Hours, Not Minutes

Short API calls are easy. Agents that work autonomously for hours need different architecture: state management, checkpoints, and graceful failure recovery.

Build vs Buy AI: How to Make the Right Call

Off-the-shelf AI tools are tempting. Custom solutions are expensive. Here is a framework for deciding which approach fits your specific situation.

The Think Tool: Teaching Your AI When to Pause and Reason

One simple trick improved our AI agent accuracy by 30%: giving it a tool to think before acting. Here is the pattern Anthropic discovered and how we use it.

Parallel AI: Running Multiple Agents on One Task

Anthropic built a C compiler with parallel Claudes. Here is how task splitting, parallel execution, and result merging work in practice.

Why Your AI Benchmarks Are Lying to You

Vendor benchmarks show what vendors want you to see. Here is how to build evaluations that actually predict whether AI will work for your specific use case.

AI Regulation in Switzerland: A Practical Guide for 2026

Switzerland is not copying the EU AI Act. Instead it is building a lean, sector-specific framework on top of existing data protection law. Here is what Swiss businesses actually need to do right now.

Running a Software Company Solo with AI: What Actually Works

A practical breakdown of which AI tools genuinely save time when you run a one-person software consultancy, which ones are overhyped, and what a real AI-assisted workday looks like.

Five AI Automations That Actually Save Time (And Three That Don't)

We tested dozens of AI automations across client projects. Here are the five that consistently deliver real time savings, the three that waste more time than they save, and the numbers behind each.

10 Strategies to Immediately Improve Your Website

Ten concrete strategies to improve your website right now, from speed and SEO to CTAs and trust signals.

How to validate your tech business idea

How to validate a tech startup idea using the Lean Startup framework, with a step-by-step fitness app example.

Why Digital Startups Are the Best Startups

Explore the unique advantages that digital startups offer, from low capital requirements to unparalleled scalability.

Finding Ideas for a Digital Startup

Six practical strategies for finding your next digital startup idea, from spotting pain points to studying successful companies.

Web Development Predictions for 2024

Six trends shaping web development in 2024: voice interfaces, AI personalization, low-code tools, PWAs, ethics, and Web3.

SEO 101 - Improve Your Google Ranking

Learn the crucial strategies to boost your website's visibility on Google and climb the search result ladder.

Key Learnings from "DotCom Secrets" by Russell Brunson

Discover valuable insights and strategies from Russell Brunson's influential book on mastering the art of online business.

Contact

Location
Switzerland
Working
Remote & On-site