A production quality pipeline for Metomic in three weeks

A data-security product team moved from stalled QA tooling work to a running Playwright pipeline, documented quality gates, and an operational handover.

3 weeks

To a running pipeline

1

Documented release gate

Customer-owned product delivery for ShopBack

A commerce product engagement built around running software, weekly clickable demos, usable documentation, and code delivered into the customer's organisation.

100%

Customer-owned delivery

Weekly

Clickable demonstrations

Designing Usher to say "I don't know"

The worst thing a website agent can do is answer confidently and wrongly. Usher's core design constraint is knowing where its knowledge ends — and routing to a human gracefully.

Cited

Every answer carries sources

Handoff

To a human, with context

the ownership economics of business software

A CFO-ready framework for the rent-vs-own decision: how metered pricing compounds against growth, what self-hosting actually costs in 2026, and a five-question test for what to own first.

Jul 1, 2026 · 12 min

The true cost of per-seat SaaS pricing (and how to model it)

Per-seat pricing taxes headcount growth, punishes adoption, and hides its real cost. A practical model for comparing seat-based SaaS against flat-priced, self-hosted alternatives.

Jun 28, 2026 · 8 min

Inside Vyom's routing decision: cost, capability, and failover

Every request through Vyom is scored against three questions: which models can do this, what does each cost right now, and what happens if the first choice is down.

Config

Vendor switches, not refactors

Per-feature

Cost attribution in SQL

Continuous accessibility: why annual WCAG audits keep failing

Annual accessibility audits are stale before they're read. Here's what CI-native WCAG 2.2 monitoring looks like, and why regressions — not initial findings — are the real compliance risk.

Jun 10, 2026 · 7 min

How Clearpath turns scans into an evidence trail lawyers accept

A compliance score means nothing in a legal dispute. What matters: what was scanned, when, against which standard, and what changed since. Clearpath is built around that record.

WCAG 2.2

AA criteria, every scan

Every deploy

Scanned, recorded, diffed

Multi-model routing: the cheapest AI optimization nobody ships

Most products send every request to one frontier model. Routing by capability class typically cuts LLM spend 30–60% with no quality loss. Here's the architecture.

May 30, 2026 · 8 min

How Verity turns four scanners into one verdict

Playwright, axe-core, OWASP ZAP, and Semgrep each speak a different language. Verity's job is making them agree on a single answer: safe to ship, or not.

4 → 1

Scanners, one verdict

New only

Only fresh regressions block a build

Agents need observability before they deserve autonomy

Production AI agents fail silently, expensively, and weirdly. Before granting autonomy, build the trace: every tool call, every decision point, every dollar, replayable.

May 12, 2026 · 6 min

Self-hosting in 2026: the rent-vs-own math has flipped

Containers, one-command deploys, and open licenses have collapsed the cost of running your own tools. A framework for deciding what to self-host — and what to keep renting.

Apr 22, 2026 · 7 min