AI & SOFTWARE

AI, built
for real work.

Articles from our enterprise development practice, covering AI agents, business systems, websites, data integration, and inbound support.

We focus on how AI can be built into practical systems for real workflows, not just introduced as a standalone tool.

Monochrome line art: a tilted balance scale weighing a stack of plates against a cluster of cubes

Can Claude Opus 4.8 Run the Lane? Opus vs Fable as Controller

Swapping the controller to Opus: +15.79 on the merge step alone, reversed across the full lane.

2026-07-27 Read →

Monochrome line illustration of a faceted sphere with small shapes visible inside — a metaphor for making work visible

When AI Makes Work Visible, Evaluation Design Decides Fairness

Unease about visibility doesn't come from AI adoption itself — it comes from loose evaluation design. Here is how we think about designing for fairness instead.

2026-07-26 Read →

Monochrome line drawing of a bare scaffolding structure standing on a ground line

For Small Businesses, Generative AI Success Doesn't Start with Model Selection

Before comparing model performance, most small businesses have implementation design to sort out first.

2026-07-24 Read →

Monochrome line art: a merged third object placed between a stack of plates and a cluster of cubes

Claude Code + Codex: Does Merging Beat the Best Solo Run?

Does a merged deliverable beat the best single one? The one case where it worked, and what it cost.

2026-07-23 Read →

Monochrome fine-line illustration of a fence running across an open field with a single gap in the middle, a metaphor for boundaries and guardrails

Designing Guardrails for AI Agents: Lessons from the OpenAI Presence Announcement

Using OpenAI's Presence announcement as a starting point, we look at how to design AI agent guardrails across scope, approval, and runtime enforcement—and what to settle before deployment.

2026-07-23 Read →

Monochrome line art: a row of identical squares splitting toward a funnel and a sealed container

AI Agent Comparison: Claude Code vs Codex on Real EC Data

The same e-commerce data split two AIs into opposite designs — and broke the scoring first.

2026-07-22 Read →

A monochrome fine-line illustration of stacked boards forming a single base, a metaphor for separating an AI agent's decisions by type

Designing Decision Authority for AI Agents: Lessons From China's Implementation Opinions

China's Implementation Opinions on AI agents frame decisions by whose authority makes them. We pair that with a process axis into a two-axis template for authority design.

2026-07-22 Read →

Monochrome line art: a Connect Four grid with black and white discs

Claude Code vs Codex: 100 Connect Four Games Head-to-Head

AI-written bots play 100 games under a shared referee — and merging them did not make them stronger.

2026-07-21 Read →

Monochrome illustration of a feedback-loop arrow returning into a stack of data (a metaphor for continuous improvement and improvement data)

What AI Adoption Leaves You: Continuous Improvement by Design

The value of AI adoption isn't decided by the model alone. We frame the 'improvement data' that stays with you and how to choose among three implementation patterns.