Topic
Pragmatic AI
12 articles
Foundry vs. Gemini Enterprise: Choosing an Agent Platform Without Betting the Farm
Two rebrands in eight months tell you the abstractions aren't stable yet. Why 'Foundry or Gemini Enterprise?' is the wrong question — and how to get value now while keeping your switching costs low.
July 6, 2026Your Agent Has a Confused Deputy Problem
The scary agent demo is the one that succeeds — using its own credentials, not the user's. Tool poisoning, the confused-deputy problem, and the boring controls that actually shrink the blast radius.
July 3, 2026Your AI Pilot Isn't Failing — Your Scoreboard Is
A VP told me their AI rollout was "a wash." The verdict was wrong — not because the tools are magic, but because the scoreboard measured the wrong things. Fix the instrument before you judge the pilot.
June 26, 2026Reimbursing AI Subscriptions Is Not a Tooling Strategy
Plenty of engineering orgs think they've solved AI tooling: expense anyone's individual subscription, done. It feels generous and fast. It's also a governance blind spot — you pay every bill and can introspect on none of it. Here's how to actually choose AI tooling for an org, and why the answer is a team or enterprise tier.
June 25, 2026The 90-Day Line: Why Some Teams Ship AI and Others Stall
High performers go from pilot to production in ~90 days. Everyone else takes nine months — with the same tools and models. The 3x gap is change management, not technology. Here's the fast path.
June 24, 2026Stop Writing an AI Policy. Start Running One.
Most leaders ask me how to write an AI policy. It's the wrong first question. The document is the easy part — and while you draft it, two-thirds of your team is already using AI they think you've banned. What you actually need is a standing committee and an operating cadence. Here's how to build both.
June 23, 2026Stop Picking a Model. Pick a Control Plane.
Two smart teams burned three hours debating Gemini vs. Claude — the wrong meeting. In 2026 the durable enterprise AI decision isn't the model, it's the control plane you standardize on.
June 22, 2026You Don't Have to Build AI. You Can Just Call It.
Most leaders picture AI as a project: hire data scientists, collect data, wait two quarters. But a huge share of useful AI isn't built — it's called. Google Cloud exposes a shelf of pretrained AI as plain APIs: speech-to-text and back, document intelligence, entity recognition, translation, image generation. Here are the quick wins you could ship this week.
June 20, 2026A Pragmatic Playbook for AI Adoption
Most AI projects stall not because the models are bad, but because the problem was never scoped for value. Here's a grounded framework for picking, shipping, and measuring AI work.
June 19, 2026You're Behind on AI. Here's How to Start.
A leader I'm working with came back from an industry event rattled — every peer was talking about their AI, and their company had nothing. The panic that produces leads to the worst first move. Here's how to start from zero deliberately: pick one workflow, measure it, and prove the win before you scale.
June 18, 2026How to Evaluate LLMs for Production
Benchmarks won't tell you if a model works for your task. Build a task-specific eval that reflects your real inputs, your real failure costs, and your real users.
June 10, 2026Reach for Retrieval Before You Reach for Fine-Tuning
When an LLM doesn't know your domain, fine-tuning feels like the obvious fix. Usually it's the expensive one. Here's why retrieval should be your first move — and when fine-tuning actually earns its keep.
June 5, 2026The Goldwater dispatch
No-hype cloud & AI, in your inbox.
New articles land Monday, Wednesday, and Friday. Get them delivered — no spam, unsubscribe anytime.