Dev Blog

Building systems that survive real use.

Essays on HubSpot architecture, AI workflow design, and engineering delivery for teams that care about what happens after launch, when operators, edge cases, and changing requirements put the real system on display.

Jul 1, 20263 min read

The Model Is Only One Part of the Interface

AI product quality depends on the full interface around the model: inputs, controls, evidence, state, review paths, and recovery behavior.

AIProductSystems

Jun 28, 20262 min read

Good AI Tools Preserve the Decision Trail

AI tools are easier to trust and improve when they preserve inputs, evidence, decisions, approvals, and corrections instead of only storing the final output.

AIGovernanceOperations

Jun 22, 20263 min read

When AI Should Ask Before Acting

The best AI workflows know when to pause for clarification, approval, or missing context instead of forcing a confident action from uncertain inputs.

AIOperationsProduct

Jun 19, 20263 min read

Retrieval Quality Is Product Quality

Retrieval is not plumbing in an AI product. It directly shapes whether the answer feels relevant, grounded, current, and useful.

AIDataArchitecture

Jun 14, 20263 min read

AI Agents Need Operating Boundaries

AI agents become more useful when their authority, inputs, tools, and escalation paths are defined before they start acting inside real workflows.

AIAutomationGovernance

Jun 4, 20265 min read

Prompt Changes Are Production Changes

Prompt edits can change routing, escalation, and decision quality in production, so they deserve versioning, evaluation, and rollback.

AIDeliveryOperations

Jun 1, 20266 min read

Documentation Is Part of the AI Stack

AI systems get more reliable when documentation defines the terms, states, tool behavior, and exception handling the workflow depends on.

AIDocumentationSystems

May 26, 20268 min read

Most AI Product Problems Are Context Problems

AI features improve when teams treat context assembly as product and systems design work instead of assuming the model can infer its way past noisy, stale, or incomplete inputs.

AISystemsArchitecture

May 10, 20265 min read

AI Workflows That Survive Contact With Reality

Durable AI systems win by structuring decisions, preserving evidence, and designing the workflow around the model for operators.

AIAutomationSystems

May 8, 20264 min read

Automation Needs a Human Exit Ramp

Good automation removes routine work but still gives people a clear, informed way to take over exceptions.

AutomationAIOperations