DEV Community: Agentic Architect

Do We Actually Need Fable 5? A Reality Check on Frontier AI

Agentic Architect — Sun, 12 Jul 2026 23:18:10 +0000

Now that Fable 5 is open to the public again, I decided to jump back in and start tinkering.

So, what are my conclusions after spending hard-earned cash on tokens?

Are Fable 5's responses good? Yes, of course they are.
Is it really expensive? Yes—in fact, I would argue that it is too expensive.

Granted, that is a complex claim to make, so let me break down how I am looking at it.

To figure out if the cost is justified, the real question you have to ask is: What is your end goal?

Are you building a full front-to-backend web or mobile app?

Are you tackling a complete website redesign?

Or do you just need it to scour your entire codebase to track down a tricky security hole or a difficult bug?

Depending on your answer, that "too expensive" label can definitely shift. I would argue that it is only truly worth spending the extra money on a frontier model if you are dealing with a massively difficult bug or a major architectural redesign. And honestly, even then, it is still up for debate.

The reality is that there are so many cheaper, highly capable models available right now that paying a premium for the absolute cutting-edge is often unnecessary.

I liken it to the evolution of smartphones. Smartphone cameras used to get noticeably better with each new release. But a few years ago, they hit a critical mass. The advancements in hardware and software between the latest flagship phone and the previous version became virtually undetectable to the human eye. The cheaper, slightly older model simply became "good enough" for almost every picture you would ever need to take.

I believe we have already reached that exact point with LLMs.

In my day job this week, I saw the release of GPT-5.6. I tested it out, and while it was undeniably good, I quickly found myself returning to GPT-5.5. The responses from 5.5 were more than good enough for the size of the codebase and the specific tasks I was working on. Plus, it used far fewer tokens, which saved my daily quota.

So in short: sure, give the frontier models a try (with Fable 5 being the current star of the show). But use them with caution and strategy. More often than not, there is another LLM out there that will do exactly what you need for a fraction of the cost.

What are your thoughts on this? I would be interested to hear how you balance cost and capability in your own workflows.

This week in Cursor + .NET — 8 rules + 1 essay (week ending July 12, 2026)

Agentic Architect — Sun, 12 Jul 2026 11:31:28 +0000

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.

Model	Time	API cost	Output quality
Claude Fable 5	about 9 min	over $10.00	Clean plan, usable first pass at the code
GLM 5.2	about 17 min	$2.76	Equally good plan and first pass for this task

	Claude Fable 5	GLM 5.2
Maker	Anthropic	Z.ai (Zhipu)
Released	June 9, 2026	June 13, 2026
Tier	Mythos-class (above Opus)	Open weights, MIT licence
Architecture	Closed	Sparse MoE, about 40B active / 753B total
Context window	1M tokens	1M tokens
Max output	128K tokens	128K tokens
Model ID	claude-fable-5	GLM-5.2 (Z.ai API / HF weights)

Per 1M tokens	Claude Fable 5	GLM 5.2	Ratio
Input	$10.00	$1.40	about 7x
Output	$50.00	$4.40	about 11x
Cached input	$1.00	$0.26	about 4x
Batch input	$5.00	(vendor batch)	n/a
Batch output	$25.00	(vendor batch)	n/a

What reviewers see	What production sees
Clean diff, idiomatic C#	Tracking overhead, stale scoped state, ignored cancellation
Green CI	Incidents that are hard to trace back to a specific PR
"We have analyzers"	Warnings buried in the IDE or not enforced on merge

Tool / plan	Price	What you actually get
Cursor Pro	approx £16/mo ($20; £12.60/mo billed annually)	$20 of included API usage, unlimited Tab completions, unlimited Auto mode. Monthly credit pool with overage.
Cursor Ultra (top tier)	approx £158/mo ($200; £126/mo annually)	$400 of included API usage and roughly 20x the Pro agent limits. Same frontier model access as Pro.
GitHub Copilot Pro	approx £8/mo ($10)	Code completions and Next Edit Suggestions stay unlimited and free. 1,500 monthly AI credits (1,000 base + 500 flex) cover agent mode, chat, CLI and PR review.
GitHub Copilot Pro+	approx £31/mo ($39)	7,000 monthly AI credits (3,900 base + 3,100 flex) and wider model choice. Completions are still free and unlimited.
GitHub Copilot Business	approx £15/user/mo ($19)	Org policy controls, audit and seat management pooled across the org, plus IP indemnity on the business plan.

Task	Reach for	Why
Inline completion while typing in Visual Studio	Copilot	Native IDE integration, already licensed
Add the same field + constructor + DI registration across 12 services	Cursor	Agentic cascade in one prompt; Copilot edits line by line
Port a controller to Minimal API + tests + run them	Cursor	Iterate-until-tests-pass is Cursor's home turf
Org-wide policy, audit, IP indemnity	Copilot Business	Org policy controls and indemnity on the business plan
Pin house style so the AI stops regressing it	Cursor	Directory-scoped `.mdc` rules + Learning Log

DEV Community: Agentic Architect

Do We Actually Need Fable 5? A Reality Check on Frontier AI

This week in Cursor + .NET — 8 rules + 1 essay (week ending July 12, 2026)

New essays this week

GLM 5.2 vs Claude Fable 5: agentic coding at a fraction of the cost

8 daily senior rules

Rule 10: AsNoTracking for Reads

Rule 9: Scoped Capture in Singleton

Rule 8: CancellationToken Propagation

Rule 7: IHttpClientFactory Discipline

Rule 6: TimeProvider Injection

GLM 5.2 vs Claude Fable 5 — the full comparison

Rule 5: Minimal API Groups

Rule 4: Validator Enforcement

Try one rule before you trust the whole kit

The AI Senior Dev Dilemma: Am I Coding or Just Prompting?

GLM 5.2 vs Claude Fable 5: agentic coding at a fraction of the cost

This week in Cursor + .NET — 7 rules (week ending July 05, 2026)

7 daily senior rules

Rule 3: Structured Logging

Rule 2: Persistence Boundaries

Rule 1: The Result Pattern

Rule 25: ActivitySource for OpenTelemetry

Rule 24: ValueTask Only When Justified

Rule 23: No Bool Flag Parameters

Rule 22: BackgroundService Over Task.Run

Try one rule before you trust the whole kit

This week in Cursor + .NET — 6 rules (week ending June 28, 2026)

6 daily senior rules

Rule 21: Channels for Producer Consumer

Rule 20: Source-Generated JSON Serialisation

Rule 19: NetArchTest for Boundaries

Rule 18: WebApplicationFactory for Integration Tests

Rule 17: ConfigureAwait false in Libraries

Rule 15: Records for Value Objects, Classes for Entities

Try one rule before you trust the whole kit

This week in Cursor + .NET — 7 rules (week ending June 21, 2026)

7 daily senior rules

Rule 14: Sealed By Default

Rule 13: Strongly-Typed IDs

Rule 12: IOptionsSnapshot Over Raw Config

Rule 11: Rethrow, Don't throw ex

Rule 10: AsNoTracking for Reads

Rule 9: Scoped Capture in Singleton

Rule 8: CancellationToken Propagation

Try one rule before you trust the whole kit

This week in Cursor + .NET — 7 rules (week ending June 14, 2026)

7 daily senior rules

Rule 7: IHttpClientFactory Discipline

Rule 6: TimeProvider Injection

Rule 5: Minimal API Groups

Rule 4: Validator Enforcement

Rule 3: Structured Logging

Rule 2: Persistence Boundaries

Rule 1: The Result Pattern

Try one rule before you trust the whole kit

The AI-generated C# that passes review and breaks in production

The pattern

Five regressions I keep seeing

1. EF Core read paths without AsNoTracking()

2. Captive dependency (scoped service in a singleton)

3. Dropped CancellationToken

4. Swallowed exception

5. Task.Run over blocking I/O

Why review misses them

Why IDE warnings aren't enough (maybe)

Two questions for you

This week in Cursor + .NET — 7 rules + 1 essay (week ending June 07, 2026)

New essays this week

Cursor vs GitHub Copilot for C#/.NET in 2026: which to pay for

7 daily senior rules

Rule 25: ActivitySource for OpenTelemetry

Rule 24: ValueTask Only When Justified

Rule 23: No Bool Flag Parameters

Rule 22: BackgroundService Over Task.Run

Rule 21: Channels for Producer Consumer

Rule 20: Source-Generated JSON Serialisation

Rule 19: NetArchTest for Boundaries

Try one rule before you trust the whole kit

Cursor vs GitHub Copilot for C#/.NET in 2026: which to pay for

1. EF Core read paths without `AsNoTracking()`

3. Dropped `CancellationToken`

5. `Task.Run` over blocking I/O