Before You Automate a Workflow: 7 Questions That Decide BUILD, REPAIR_FIRST, DEFER, or STOP
Seven questions that decide whether one workflow should be automated now, repaired first, deferred, or left alone before you commit budget to an AI build.
First AI Movers — Article Archive
941 original articles by Dr. Hernani Costa, PhD, on AI strategy, EU AI Act compliance, governance, and agentic systems for European businesses. Published under CC BY 4.0 — free to read, cite, adapt, and redistribute with attribution.
941articles
103canonical topics
February 2025first article
September 2026most recent
Latest articles
Choose the Reference That Tests the Decision
Five tests that decide whether a dataset, benchmark or prior deployment can carry the decision you have to make, before you build an evaluation on it.
CRA Reporting Readiness: The Platform Field Is Not the Gap
The CRA platform already records when you became aware. The readiness gap is internal: awareness evidence, named ownership and a clock you can defend.
0.992 Was Not Replication: Why One Exciting Posterior Is Not a Causal Story
A T-cell signal reached PP4 near 0.992 in one dataset. Three independent public blood T-cell resources did not reproduce it.
Try to kill your target before you fund the next expensive step.
A fixed-scope, five-business-day adversarial review of one biological target or mechanism, ending in an ADVANCE, REWORK, STOP or NOT\_EVALUABLE memo.
Biodius Target Challenge FAQ: Scope, Inputs, Price and Limits
What the Biodius Target Challenge reviews, what you send, what you get back, what it costs, and what it does not claim to do.
Falsification-First AI for Biological Target Assessment: What Four Prospective Experiments Taught Us About Scientific Agents
A technical report on four prospective experiments: why scientific AI is more useful when it preserves negative evidence and stops when data runs out.
Not Evaluable Is a Result: When Public Human Genetics Cannot Decide the Causal Bridge
A plausible mechanism survived AI review. Public human genetics still could not decide the causal bridge, so the hypothesis was not promoted.
Why We Built a Service That Tries to Kill Your Biological Target
Four prospective experiments taught us that scientific AI is most useful when it attacks a hypothesis. So we built a fixed-scope target diligence review.
July 2026 Momentum Across Open Agent Platforms
Visible momentum across open agent platforms in July 2026, with new releases from Dify and LangChain. What this means for European engineering teams.
Open-Source Coding Assistants and Autonomous Agents: The New State of Software Development
Open-source coding tools surge: AI editors and autonomous agents from independent maintainers. A guide for European engineering leaders.
We Built the Engine But Not the Chassis: What AI-Accelerated Teams Are Not Telling You
AI gave small teams infinite velocity but broke their processes. The data on cognitive debt, brain fry, and what surviving teams do differently.
The Honest AI Adoption Guide: Every Fear You Have Is Valid, Here Is What to Do About Each One
You are busy, sceptical, and worried about wasting money on AI. This guide addresses every real fear European SME leaders have about AI adoption, honestly.
AI Engineering for European SMEs: If You Can Describe It, We Can Build It
First AI Movers builds custom AI platforms, SaaS products, and automation systems for European SMEs, from idea to production on any cloud or model, with no vendor lock-in.
A Two-Person Team Out-Built 300 Rivals in 12 Hours, Because They Mapped First
Most failed AI pilots die from no plan, not bad tech. Learn the map-first AI project planning method that lets small European teams ship working AI.
Content Provenance: A Checkable History That Travels With Every File
Content provenance attaches a tamper-evident record to media, so you can verify origin and edits instead of trusting appearances. C2PA and open tools make
Confidential Computing: Protecting Data While It Is In Use
Confidential computing protects data during processing with hardware-based trusted execution environments, closing the gap between encryption at rest and
What Makes an AI System Trustworthy: The Characteristics NIST Names
NIST's AI RMF names seven characteristics of trustworthy AI: validity, safety, security, accountability, explainability, privacy, and fairness
How the Sentry Intercepts System Calls to Contain Untrusted Agent Code
The Sentry in gVisor mediates every system call from untrusted code, reducing the host attack surface. Learn how interception works and why it matters for
Secure by Design: Safe Defaults for Teams Building AI Agents
Apply CISA Secure by Design to AI agents: make safe configurations the out-of-the-box default, so teams ship agents that are secure before anyone touches
Machine-readable surfaces
/index.json— catalog of all 941 articles with metadata/feed.xml— Atom feed of the 50 most-recent articles/feed.json— JSON Feed 1.1 of the 50 most-recent articles/llms.txt— LLM discovery summary/llms-full.txt— full-text corpus for bulk LLM ingestion (941 articles, ~7 MB)/llms-recent.txt— rolling 30-day window for small-context LLM ingestion/sitemap.xml— XML sitemap pointing to canonical article URLs/hernanicosta.json— Schema.org Person entity for the author