17th Street Labs BlogPractical AI engineering lessons: LLM evals, agents, model costs, local models, and AI security.https://17thstreetlabs.com/en-usNew Models, 3 New Evals: OCR, Translation, and Security Verification Testshttps://17thstreetlabs.com/blog/holy-model-release-week/https://17thstreetlabs.com/blog/holy-model-release-week/GPT-6, Claude, Gemini, GLM, DeepSeek and Qwen on our handwriting OCR, translation and security-verifier tests—with the costs, ties and judge disagreements left in.Thu, 24 Sep 2026 19:32:48 GMTDan LevyTesting & EvaluationSmall Models. Serious Work.https://17thstreetlabs.com/blog/small-local-models/https://17thstreetlabs.com/blog/small-local-models/What small local models can do, where they struggle, and how to give them a fair test.Mon, 14 Sep 2026 22:07:01 GMTMarina LevyDan LevyAgentsThe Cheap Model Can Get Expensive.https://17thstreetlabs.com/blog/cost-per-completed-task/https://17thstreetlabs.com/blog/cost-per-completed-task/The number worth watching is cost per completed task. Here’s how to measure it.Mon, 14 Sep 2026 22:07:01 GMTMarina LevyDan LevyCosts and OptimizationBefore You Buy the GPU.https://17thstreetlabs.com/blog/renting-vs-buying-gpus/https://17thstreetlabs.com/blog/renting-vs-buying-gpus/Renting compute can buy you something more useful than hardware: room to change your mind.Mon, 14 Sep 2026 22:07:01 GMTMarina LevyDan LevyCosts and OptimizationOne Harness Does Not Fit Every Model.https://17thstreetlabs.com/blog/one-harness-does-not-fit-every-model/https://17thstreetlabs.com/blog/one-harness-does-not-fit-every-model/The structure that helps a small model can get in a stronger model’s way.Mon, 14 Sep 2026 22:07:01 GMTMarina LevyDan LevyAgentsThe Attackers Are Getting Agents, Too.https://17thstreetlabs.com/blog/continuous-security-testing/https://17thstreetlabs.com/blog/continuous-security-testing/Cheap automation changes the pace. Your security testing needs to keep up.Mon, 14 Sep 2026 22:07:01 GMTMarina LevyDan LevySecurity & PrivacyYour AI Judge Needs to Be Judged, Too.https://17thstreetlabs.com/blog/eval-driven-agent-development/https://17thstreetlabs.com/blog/eval-driven-agent-development/A practical approach to eval-driven development: test the journey, then test the grader.Mon, 14 Sep 2026 22:07:01 GMTMarina LevyDan LevyTesting & EvaluationYour Tests Passed. Your Interface Didn’t.https://17thstreetlabs.com/blog/agents-that-see-the-ui/https://17thstreetlabs.com/blog/agents-that-see-the-ui/Give an agent the browser recording, and it can look for what the final screenshot missed.Mon, 14 Sep 2026 22:07:01 GMTMarina LevyDan LevyTesting & EvaluationSome Data Doesn’t Get to Leave.https://17thstreetlabs.com/blog/local-ai-data-privacy/https://17thstreetlabs.com/blog/local-ai-data-privacy/Local AI can keep sensitive work inside your environment. Check the whole route.Mon, 14 Sep 2026 22:07:01 GMTMarina LevyDan LevySecurity & PrivacyThe Agent Is Working. Does Anyone Know What It’s Doing?https://17thstreetlabs.com/blog/security-copilot-that-explains/https://17thstreetlabs.com/blog/security-copilot-that-explains/A security copilot should make the work easier to follow—and easier to question.Mon, 14 Sep 2026 22:07:01 GMTMarina LevyDan LevySecurity & Privacy