→ Engineering notes
How we build AI that ships to production.
Cost models, architecture patterns, eval, observability, and the integration work that decides whether an AI feature ever reaches real users. Written for engineers and the people who buy engineering.
- ComplianceOct 9, 2026 · 9 min
EU AI Act Compliance Checklist for Teams Shipping AI
Most of the EU AI Act is engineering work. Classify every use case, ship the Article 50 disclosures that have applied since August 2026, and log what you cannot reconstruct later.
Read post
- StrategyOct 9, 2026 · 9 min
AI Readiness Checklist: 25 Questions Before You Build
Maturity models tell you where your company sits. This checklist tells you whether one use case is ready to build: 25 questions, a score, and what it means.
Read post
- StrategyOct 9, 2026 · 9 min
AI Vendor Evaluation Checklist: Scoring Tools and Partners
Nine criteria, weighted before the first demo, for scoring an AI product or an AI development partner. Plus the questions to send in writing and the answers that end the evaluation.
Read post
- ComplianceOct 9, 2026 · 9 min
Is My AI Chatbot High Risk Under the EU AI Act? Usually Not
A support or knowledge chatbot is limited risk, with disclosure duties since August 2026. It turns high risk when its output feeds hiring, credit, insurance or education decisions.
Read post
- PricingOct 9, 2026 · 8 min
What AI Consulting Costs a Small Business in 2026
Hourly advice, a fixed fee assessment, a fractional lead or a build partner. What each one costs a small business, what you should walk away with, and how to tell who has actually shipped.
Read post
- StrategyOct 9, 2026 · 8 min
AI Feasibility Study: What It Should Contain and Cost
Data, technical, economic and compliance feasibility, what the study should cost, how it differs from a PoC, and how to spot one written to close a sale.
Read post
- ComplianceOct 9, 2026 · 9 min
AI Governance Framework for a Company of 50 to 750 People
AI governance is a register, named owners and a few written rules, not a committee. This is the eight-part version we recommend for companies of 50 to 750 people.
Read post
- StrategyOct 9, 2026 · 9 min
How Can AI Help Small Businesses? What Works, by Function
Support, sales, finance, operations and internal knowledge. What AI does well in each today, whether a tool or a custom build fits, roughly what it costs, and when AI is the wrong answer.
Read post
- ComplianceOct 9, 2026 · 10 min
EU AI Regulations Explained for Software Teams in 2026
The AI Act is only one of the EU rules that apply to AI software. How it sits next to GDPR, the Data Act, NIS2 and DORA, with the dates as amended in July 2026.
Read post
- StrategyOct 9, 2026 · 9 min
Build vs Buy AI: A Decision Framework With Three Year Costs
Buy what is common, build what is yours, and own whatever holds your data. A decision matrix, a three year cost comparison and the middle path most B2B companies end up on.
Read post
- ToolingJun 8, 2026 · 7 min
The Best AI Tools for B2B SaaS in 2026 (That Engineering Teams Actually Use)
Forget the 100-tool 'AI awesome' lists. Here's the tight stack we actually ship into B2B SaaS builds in 2026, picked because they survive production, not because they look good on Twitter.
Read post
- ComplianceJun 5, 2026 · 8 min
GDPR-Compliant AI Deployment for EU Companies: A Practical Engineering Guide
GDPR + AI is solvable. Most teams get it wrong by accident: a sub-processor not disclosed, data crossing borders silently, prompts logged where they shouldn't be. Here's the discipline.
Read post
- IntegrationJun 4, 2026 · 8 min
AI Integration with SAP: Practical Patterns for ERP-Native AI Features
SAP integrations don't have to take a year. Here's how we ship AI features against S/4HANA and Business One in 6-10 weeks.
Read post
- IntegrationJun 3, 2026 · 8 min
AI Integration with Salesforce: A Practical Guide for Engineering Teams
Salesforce has its own AI stack. You don't have to use it. Here's how to build custom AI features against the Salesforce APIs you already have.
Read post
- InfrastructureJun 2, 2026 · 9 min
How to Deploy an LLM On-Premise in 2026 (Llama 3 on vLLM Guide)
Sovereign LLM deployments are no longer rocket science. Here's the modern stack (Llama 3 70B on vLLM behind a private API) and what it actually takes to operate.
Read post
- EngineeringJun 1, 2026 · 8 min
Why AI Projects Fail in Production (And What Engineering Teams Miss)
Most AI projects that fail share the same fingerprints: no eval, prompt sprawl, cost surprises, integration that never lands. Here's the full failure-mode catalog.
Read post
- Buyer guideMay 29, 2026 · 6 min
How to Write a Brief for an AI Project (That Gets a Useful Quote Back)
Most AI quotes come back vague because the brief was vague. Here's the 8-section template we wish every prospect would send.
Read post
- StrategyMay 28, 2026 · 6 min
AI Proof-of-Concept vs MVP: What's the Difference, and Which Do You Need?
Most failed AI projects ship a POC and call it an MVP. The gap between them is where the eval suite, the integration, and the runbook live.
Read post
- StrategyMay 27, 2026 · 7 min
Five Signs You Need an AI Consultant (And Two Signs You Don't)
Most CTOs bring in AI help too late, after the prototype has quietly stalled. Here is how to spot the moment, and what a good AI consultant should hand you.
Read post
- PricingMay 26, 2026 · 7 min
How Much Should You Pay an AI Development Agency in 2026?
If you're buying AI development services, here's the bracket you should expect and the questions to ask if a quote falls outside it.
Read post
- EngineeringMay 25, 2026 · 7 min
AI Integration Checklist for Engineering Teams (2026)
Twenty-six items we work through before shipping AI to production. Run this against any AI project you're scoping this quarter.
Read post
- IntegrationMay 22, 2026 · 9 min
How to Add AI to an Existing SaaS Product Without Breaking It
Don't bolt AI onto the side. Most failed integrations are the ones that became a separate UI users never adopted. Here's how we wire AI inside the product.
Read post
- StrategyMay 21, 2026 · 8 min
Open-Source vs Closed-Source LLM for Business: A 2026 Decision Guide
Closed-source frontier models win on day one for most teams. Open-weights win when data residency, customization, or unit economics force the move.
Read post
- OperationsMay 20, 2026 · 9 min
How to Evaluate an LLM in Production: A Practical Eval Guide
Most teams skip eval until the model breaks in production. By then they cannot tell why. Here's the eval pipeline we build into every project.
Read post
- OperationsMay 19, 2026 · 8 min
AI Cost Optimization in Production: Where the Money Actually Goes
Token spend is the part most teams discover too late. Here are the seven moves we apply on every production AI build to keep unit economics healthy.
Read post
- ArchitectureMay 18, 2026 · 7 min
How to Choose a Vector Database for AI Search in 2026
Most teams overthink this. pgvector on the Postgres you already operate handles 95% of B2B use cases. Here's when the other options actually win.
Read post
- PricingMay 18, 2026 · 9 min
How Much Does It Cost to Build an AI Agent in 2026?
Most production AI agents we scope land between €25,000 and €80,000. Below €10,000 you are usually buying a tool, not building an agent. Here is how the number gets made.
Read post
- ArchitectureMay 15, 2026 · 8 min
Claude vs GPT-4o vs Open-Weights LLMs: Picking the Right Model in 2026
The right LLM depends on the task, not the brand. Here's the decision tree we use when scoping production AI builds for B2B teams.
Read post
- InfrastructureMay 15, 2026 · 8 min
Lambda Labs for AI Products in 2026: When Your AI App Needs Dedicated GPU Infrastructure
Lambda Labs gets mentioned the moment a team thinks about GPUs. The real question is whether the AI product actually needs dedicated GPU infrastructure, or whether managed APIs are still the right call.
Read post
- StrategyMay 14, 2026 · 7 min
How Long Does It Take to Build an AI MVP in 2026?
Three weeks from kickoff is realistic for a focused scope. Six is realistic for a real production system. Here's what each week actually contains.
Read post
- ArchitectureMay 13, 2026 · 9 min
RAG vs Fine-Tuning in 2026: A Decision Framework for Production AI
Almost every team gets this call wrong. The default answer is RAG, and fine-tuning is the last resort. But there are specific cases where the opposite is true.
Read post
Want this applied to your project?
Read the notes, then send the brief. We come back with a preliminary scope and a budget bracket within one business day.