Read the full analysis: The Practical Guide To Choosing AI For Software Development on ThorstenMeyerAI.com
TL;DR
This article offers a detailed framework for selecting AI models in software development, emphasizing matching models to tasks and effort levels. It highlights five key models—Sol, Luna, Astra, Opus, and Fable—and explains their specific roles. The guide aims to help teams optimize AI use, reduce costs, and improve reliability.
A new comprehensive guide has been released to help software development teams select the appropriate AI models for various tasks, emphasizing the importance of matching models to specific effort levels and requirements. The guide, based on insights from Thorsten Meyer, details five AI models—GPT‑6 Sol, Luna, Astra, Opus, and Fable—and their optimal use cases, aiming to improve efficiency and reduce costs in AI-assisted development.
The guide categorizes AI models according to their strengths and ideal applications, advocating for a structured approach rather than a one-size-fits-all solution. GPT‑6 Sol is recommended for routine implementation work, such as features, UI, and API tasks, where clear interfaces and acceptance criteria are present. Luna is suited for bounded, repeatable tasks like documentation, translation, and testing, where reliability and inexpensive checks are critical. Astra and Fable are reserved for demanding reasoning tasks and complex, multi-step development, respectively, with Astra handling architecture, security, and complex logic, and Fable managing extended, intricate packages.
Opus serves as an independent reviewer or for implementation requiring a second perspective, especially in critical review stages. The guide emphasizes that each model should be paired with a verification step—without which, the recommendation remains a guess. The lifecycle table in the guide pairs tasks, models, effort levels, and required checks, underscoring the importance of verification to avoid costly mistakes.
Thorsten Meyer highlights that misusing models—such as applying a single model for all work or neglecting effort calibration—leads to inefficiencies and errors. The guide aims to help teams allocate AI resources more effectively, balancing cost, effort, and accuracy, ultimately leading to more reliable software development processes.
DEVELOPMENT · MODEL & EFFORT GUIDE
A practical guide to AI‑assisted development
Sol for implementation, Luna for bounded routine work, Astra and Fable for demanding reasoning, and Opus for implementation or a second perspective. Use a clear contract and observed evidence throughout delivery.
Escalate the uncertainty, not the effort
A second perspective at any level: a separate review task with explicit adversarial questions.
When you escalate, hand over the failing case and the evidence, not “try harder.” Astra and Fable can review each other’s work, with separate files and independent acceptance evidence.
What each model is for
Complex decisions
GPT‑6 Astra
Architecture, security boundaries, difficult debugging, data migrations, distributed behavior, multi‑system integration.
High for consequential changes; Extra High for unresolved, interacting constraints.
Everyday implementation
GPT‑6 Sol
Features, UI and API work, refactoring, meaningful tests, automation, bug fixes within a defined scope.
Medium as the working default; High for complex logic and cross‑module changes.
Focused execution
GPT‑6 Luna
Documentation from evidence, structured extraction, small mechanical edits, translation checks, fixed test scripts.
High as a starting point. Escalate permissions, business meaning or destructive operations.
Implementation & independent review
Claude Opus 5.5
Can own a bounded implementation package; especially useful as a separate reviewer challenging another agent’s assumptions and tests.
Medium for well‑defined implementation; High for critical reviews.
Demanding extended development
Claude Fable 5.1
Complex packages spanning many steps, architectural investigations, or a deep independent review.
High as a starting point, with checkpoints and a usage budget.
Verify which effort settings your client and account actually offer.
Allocate work across the lifecycle
Requirements and scopeSol Medium; Astra High for ambiguityExamples, exclusions, unresolved decisions, acceptance criteriaArchitecture and public contractsAstra HighAlternatives, failure modes, compatibility, independent reviewUI, accessibility and localizationSol MediumReal interaction, keyboard use, relevant languages and screen sizesBusiness logic and API implementationSol High for complex workPublic‑interface tests, validation, errors and retriesAuthentication and tenant isolationAstra High / Extra HighNegative cross‑tenant, role, session and object‑access tests; independent reviewDatabase migrations and concurrencyAstra HighReal database, contention, failed transactions, restore and rollbackSmall mechanical refactorsLuna High or Sol MediumDiff review and a focused regression checkDifficult or intermittent defectsSol High → Astra High if unresolvedReproduction, hypothesis, isolated cause, regression testFixed browser / device acceptanceSol Medium; Luna for recordsActual target device/browser and exact build identityBenchmark and evaluator designAstra High or Fable High + independent reviewerIndependent oracle, held‑out cases, meaningful thresholds, no target‑score tuningExtended multi‑module developmentFable High or Astra High; Sol for bounded subtasksMilestone evidence, fixed interfaces, one integration owner, independent reviewDeployment and production recoveryAstra High for planning and high‑risk changesBound artifact, actual target, backup/restore, health checks, authorized rolloutRelease notes and maintenance recordsLuna HighTrace every claim to executed evidence; Sol checks completeness
One delivery workflow, clear ownership
1
Outcome, scope, interfaces, acceptance tests, budget and stop conditions. Read repository instructions first.
2
Bounded packages, distinct files, one integration owner. Parallelize only independent work.
3
Authorization, loading, empty states, failure, cancellation, retry, recovery. Preserve unrelated changes.
4
Public entry points and real dependencies. Keep simulated results separate from real evidence.
5
Counterexamples and dangerous failure directions, with independently derived expectations.
6
Validate the combined artifact, migrations and recovery path. Passing tests are not approval.
7
Check the deployed version and critical flows. Record limits, signals, ownership, follow‑ups.
Four rules that prevent expensive mistakes
Reusable task brief
Outcome: [observable user or system result]
Scope: [included work and explicit exclusions]
Contract: [repository instructions, plan, interfaces]
Ownership: [allowed files; integration owner]
Model / effort: [recommendation and reason]
Acceptance: [real flows and objective success criteria]
Negative cases: [permissions, stale data, retry, concurrency]
Evidence: [commands, outputs, artifact/build identity]
Constraints: [time/credit budget, dependencies, data boundaries]
Escalation: [uncertainty that requires review or user input]
Release: [destination, authorization, migration and rollback]
Finish: [reviewable changes, test evidence, limits, next steps]
ThorstenMeyerAI.comGuide only: no model configuration or deployment changes. Model roles are informed by vendor documentation (OpenAI · Models & reasoning effort, Anthropic · Models overview). The allocation is an engineering recommendation, not a measured ranking or a guarantee of safety; validate it on your own codebase. Updated 23 September 2026.
Why Proper AI Model Selection Improves Development Efficiency
Choosing the right AI model for each task can significantly reduce development costs, improve code quality, and minimize errors. By matching models to specific effort levels and verification needs, teams can avoid wasteful spending on routine work or costly mistakes in complex decision areas. This structured approach enhances trust in AI outputs and streamlines workflows, which is especially vital as AI becomes more integrated into software engineering.
Background of AI Use in Software Development
The adoption of AI in software development has grown rapidly, with models like GPT-6, Claude, and Fable becoming integral tools. Early efforts focused on replacing manual coding or automating simple tasks, but as capabilities expanded, the need for structured guidance on model selection emerged. Previous approaches often applied a single AI model across multiple tasks, leading to inefficiencies and errors. This guide builds on recent insights from Thorsten Meyer, emphasizing task-specific model deployment aligned with effort and verification levels to optimize outcomes.
“Most teams using AI for software development make the same two mistakes: they pick one model for everything, and they solve every hard moment by turning the effort up. Both waste resources and increase risk.”
— Thorsten Meyer
Unresolved Questions About Model Effectiveness and Application
While the guide offers a structured approach, it is not yet clear how well these recommendations perform across different team sizes, project types, or in real-world production environments. The effectiveness of effort calibration and verification steps in practice remains to be validated through broader adoption and empirical testing. Additionally, the evolving capabilities of AI models may shift optimal use cases over time, requiring ongoing adjustments.
Next Steps for Teams Implementing AI Model Frameworks
Organizations are encouraged to pilot the recommended model-task pairings within their projects, monitor outcomes, and refine their approach based on results. Further research and case studies are expected to emerge, validating and potentially expanding the framework. As AI models continue to improve, updates to the guide may incorporate new features and best practices, making ongoing education and adaptation essential for teams seeking to maximize AI benefits.
Key Questions
How do I determine which effort level to assign to a task?
Effort levels are based on task complexity, uncertainty, and importance. Routine, well-understood work typically requires lower effort settings, while complex or uncertain tasks need higher effort and verification. The guide provides a lifecycle table to help match effort levels with specific task types.
Can I use a single AI model for all tasks in my project?
No. The guide emphasizes that different tasks require different models and effort levels. Using one model for everything risks inefficiency and errors. Tailoring models to task specifics improves accuracy and cost-effectiveness.
What verification steps are recommended for ensuring AI output quality?
Verification depends on task type but generally includes independent reviews, negative testing (e.g., security checks), and traceability of outputs to actual executed evidence. The guide stresses pairing each model recommendation with a specific check to confirm correctness.
How often should I revisit my AI model deployment strategy?
Regular review is recommended as AI capabilities evolve and project needs change. Ongoing monitoring of outcomes and incorporating feedback will help adapt the framework for continuous improvement.
Source: ThorstenMeyerAI.com