Why Teams Should Test AI Tools Before Spec Sessions, Not After

·1 views

A new essay on DEV Community draws a parallel between ancient Athenian ephebeia training and how modern teams should build calibration with AI agents before committing to a specification. The author argues that most teams skip directly to planning sessions without first developing shared, hands-on experience of what an AI tool can and cannot do. This leads to Spec Sessions dominated by untested assumptions — some overconfident, some overly cautious — depending on who speaks first. Just as young Athenians trained together in gymnasia under real conditions before taking civic roles, the author contends that teams should run AI tools against real tasks in low-stakes settings to build genuine collective judgment. The core argument is that understanding a tool and testing a tool are not separate activities — the testing is how the understanding gets built.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

Ego Development Model Maps Eight Stages of How Professionals Construct Meaning

Researcher Susanne Cook-Greuter built on Jane Loevinger's 1970s framework to empirically map eight ego development stages — from Self-Centric to Unitive — using the Sentence Completion Test instrument. The model describes how individuals organize experience and construct meaning, rather than measuring personality or knowledge. Two professionals with identical skills can respond to the same situation in fundamentally different ways depending on their developmental stage. The framework covers mechanisms such as stage transitions, regression under stress, and a recognition asymmetry where higher stages can understand lower ones but not vice versa. The article is part of a broader Peopleware series applying ego development theory to software engineering and organizational contexts.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

How to Debug Low-Level Design When Your Domain Model Feels Off

Software engineers often encounter domain models that compile correctly but feel structurally wrong, with symptoms like scattered responsibilities, duplicated logic, and unclear ownership. Common root causes include anemic domain models where entities hold no behavior, fat services that absorb all business logic, and poorly defined aggregates that blur lifecycle boundaries. Hidden coupling between modules and uncontrolled state transitions further weaken design integrity, making systems fragile and hard to maintain. Experts recommend tracing issues back to the real user journey and asking who owns each business rule, rather than layering on more classes or abstractions. Strong domain modeling is ultimately about aligning code structure with actual business behavior, ensuring that invariants, boundaries, and state transitions reflect real-world logic.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Source View Technology Merges APT and AST to Make Java Code Enhancements Visible

A new approach called Source View Technology aims to overcome the individual limitations of two Java compile-time tools: APT (Annotation Processing Tool) and AST (Abstract Syntax Tree) modification. APT can generate observable, debuggable source files but cannot modify existing classes by producing a class with the same fully qualified name. AST modification, used by tools like Lombok, can enhance existing classes in memory at compile time but leaves those changes hidden, undebuggable, and only inspectable by decompiling bytecode. Source View Technology combines both approaches by using AST to modify classes and then outputting the result as source files, called View Classes, which coexist with the original developer-written classes. At compile time, these View Classes replace the originals, giving developers the ability to modify existing classes while keeping all enhancements fully visible and debuggable.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Developer shares hard lessons building a Slack deploy queue bot with NestJS and Redis

A software developer documented the technical challenges of building a Slack bot to manage deployment queues across multiple teams and environments, sharing lessons learned in production. The project used NestJS, PostgreSQL via Prisma, Redis with BullMQ for background jobs, and Slack's Bolt SDK for integration. One key challenge involved Slack's Block Kit lacking native support for dependent dropdowns, which was solved by dynamically updating modals and changing block IDs to bypass Slack's internal state cache. Timeout logic for queue entries was made restart-safe by storing scheduled jobs in Redis via BullMQ with deterministic job IDs, rather than relying on in-memory setTimeout calls. The developer also replaced a stored queue-position column with a runtime-computed SQL query to avoid race conditions and data corruption when users leave the queue mid-process.

0 comments Read more at DEV Community