Engineering reviews
Engineering reviews of every tool, framework, and stack we run, have run, or have evaluated in earnest. Each review is grounded in production observation. Each is updated quarterly. Each names what is broken alongside what works.
By Oliver Wakefield-Smith, Digital SignetHow we review: our methodology. The taxonomy below mirrors the structure of our pipeline: coding agents (the build layer), frameworks (the orchestration layer), autonomous & enterprise (the autonomy layer + procurement-vertical bridges).
Strongest model-grade coding agent we have used. Most expensive when unsupervised.
Composer is the best multi-file editing experience in 2026. Tab-completion is the rest of why we keep it.
Issue-to-PR mode is competent. PR-creation success is reliable on small issues, less so on medium.
Task-priced, sandbox-driven. P95 cost-per-task is where the economics get sharp.
Strong demo-to-deploy speed. Failure modes are post-prototype.
Cursor-alternative for non-engineers. The non-engineer claim is overstated.
Hosted-deploy context is the strength. Local-deploy context goes to Claude Code.
UI-generation specialist. We use it as a sub-tool inside a larger pipeline.
Reputation has shifted since 2024. Right when LangGraph would be overkill.
What we use in production. Linear scaling, explicit graph, the operator credential matters here.
Fastest path to a working role-based prototype. Hits a ceiling around 5 concurrent agents.
Conversation pattern is expressive. The conversation pattern cost economics is the catch.
AutoGPT, MetaGPT, Pydantic AI, DSPy, smolagents. One review per framework, summary tier.
+9,999,900% YoY signal. Architecture analysis, security stress-test, the rebrand history.
Full virtual computer. Real on research and scraping. Not real on app-building.
Open-source self-hostable. Strong trend signal, deployment cost is the catch.
Browser-use class. Production risks are real. Includes Anthropic Computer Use sub-section.
Computer-use agent in research workflow. Where it shines and where it does not.
Enterprise low-code. What the no-code claim costs you in flexibility.
Salesforce stack. Light editorial; procurement-grade detail at the vertical sites.
IBM enterprise agent stack. Independent technical review.
Oliver runs Digital Signet, a research and product studio that operates ~500 production sites with AI agents as the engineering layer. The Digital Signet portfolio is built using a continuous AI-agent build pipeline, one of the largest agent-operated publishing operations on the open web. The handbook draws directly from those deployments: real cost data, real failure modes, real recovery patterns.