Anastasiia Bakhtoiarova·Nov 9Why Is Enterprise AI Slower Than Startup AI, and When Is That Justified?A stalled AI project looks identical whether the cause is a real risk or a process nobody questions. The fixes have nothing in common.
Anastasiia Bakhtoiarova·Nov 5Reliability isn’t free: what each check costs in a multi-agent pipelineWhere the cost of trusting a multi-agent pipeline lands, and the check I still can’t audit.
Anastasiia Bakhtoiarova·Nov 2The Model Is Not the Product. The System Around It Is.Changing a model or a prompt is a deployment, and it should be gated like one.
Anastasiia Bakhtoiarova·Oct 13Turning “Looks Good” Into an Objective FunctionThe hardest part of building a wardrobe recommender wasn’t the agents. It was deciding, precisely enough to compute, what “good” even…
Anastasiia Bakhtoiarova·Oct 5When Your Eval Metrics ConflictA single score doesn’t resolve a disagreement between dimensions. It picks a winner and doesn’t tell you.
Anastasiia Bakhtoiarova·Sep 28Your RAG System Retrieved the Right Documents. So Why Is the Answer Still Wrong?Retrieval and reasoning are two separate claims.
Anastasiia Bakhtoiarova·Sep 9What Happens When You Try to Test Something That Doesn’t Give the Same Answer Twice?It inherits the hard parts of both and the safety net of neither — notes from building a small evaluation framework for my own agents.
Anastasiia Bakhtoiarova·Sep 1More Context Isn’t a Free Upgrade.What I learned about relevance, dilution, and conflicting instructions while building a four-agent trip planner
Anastasiia Bakhtoiarova·Aug 20The First Agent Is Easy. The Second One Changes Everything.What building a two-agent project taught me about handoffs, hidden contracts, and evaluating a system instead of a componentA response icon1A response icon1
Anastasiia Bakhtoiarova·Aug 7Prompt Engineering Is Not Engineering. It’s System Design.What I noticed in my own AI-agent experiments when a better prompt fixed the symptom but not the architecture underneath it