Every weekday a single, opinionated rule for senior C#/.NET engineers using Cursor. Here's the full week in one read — canonical posts live on the Agentic Architect blog.
New essays this week
GLM 5.2 vs Claude Fable 5: agentic coding at a fraction of the cost
Tue 07 Jul · AI models
I ran the same agent task on both. Fable 5 is faster and stronger on benchmarks, but GLM 5.2 finished for $2.76 vs over $10. The honest cost math for running agents on open weights.
8 daily senior rules
Rule 10: AsNoTracking for Reads
Sun 12 Jul
Every read-only EF Core query should call AsNoTracking. Add a rule that recognises query methods returning DTOs (not entities) and inserts the call. Cursor never does this by default and your read perf degrades silently across releases.
Rule 9: Scoped Capture in Singleton
Sat 11 Jul
The single most expensive .NET runtime bug: a Singleton holding a Scoped service. Cursor cheerfully writes this without warning. Audit constructor parameters of any class registered as Singleton — if any are typically Scoped (DbContext, repositories, MediatR sender), flag it before merge.
Rule 8: CancellationToken Propagation
Fri 10 Jul
Every async method in your codebase should accept and forward a CancellationToken. Make it a rule that any new async signature without one is flagged. Stops the AI from quietly losing cancellation half-way down a call chain.
Rule 7: IHttpClientFactory Discipline
Thu 09 Jul
Refuse new HttpClient() in any generated code. Force Cursor to inject IHttpClientFactory or a typed client. The classic socket-exhaustion bug is exactly the kind of footgun an AI hands you without realising.
Rule 6: TimeProvider Injection
Wed 08 Jul
Never let Cursor hardcode DateTime.Now or DateTime.UtcNow in business logic. Inject TimeProvider (or your own IClock) instead. This makes time deterministic for tests and stops the AI reaching for static APIs whenever it generates time-aware code.
GLM 5.2 vs Claude Fable 5 — the full comparison
Wed 08 Jul
Side-by-side: pricing per million tokens, benchmarks, context windows, and the honest verdict on which model to use for which kind of agent run. Open weights vs closed frontier, with the real cost math.
Rule 5: Minimal API Groups
Tue 07 Jul
Use MapGroup and EndPoints for .NET 9 services. It keeps the context window lean and focused, helping the AI provide more accurate code suggestions than it can with massive, bloated Controllers.
Rule 4: Validator Enforcement
Mon 06 Jul
Every MediatR Request must have a FluentValidation rule. Force the AI to check the validator before writing handler logic. This prevents "context rot" where the AI forgets your business constraints.
Try one rule before you trust the whole kit
The free arch-core-lite.mdc is one drop-in Cursor rule that ends the morning re-explanation ritual. Install in 60 seconds, see whether Cursor actually remembers your DI lifetimes, and decide for yourself whether the full kit is worth £9.00.
-
Free sample:
arch-core-lite.mdcon GitHub - Full kit (£9.00, one-time): Agentic Architect Kit
- Daily rules feed: https://agentic-architect.dev/blog/
Canonical home for everything in this digest: https://agentic-architect.dev/blog/.
Top comments (0)