6.4.2025

Conducting an AI-Native Developer Assessment (RS-DEVS)

The AI-native approach helps a PM remove subjectivity, rely on workflow facts, and prepare arguments for growth-focused feedback.

Step 1. Collect artifacts (facts base for the PM)
Step 2. Initialize the AI and load the context
Step 3. Go through the assessment stage by stage (dialog with AI)
Step 4. Build the RS-DEVS report and development plan (PDP)

Conducting an AI-Native Developer Assessment (RS-DEVS) 6/4/2025 Author:

The foundational AI-native approach helps a PM remove subjectivity, rely on facts from work processes, and prepare rock-solid arguments for growth-focused feedback.

According to company policy, the engineer's regular review is conducted by

The manager (PM) after the developer completes the self-assessment.

PMs often tend to unintentionally inflate scores because of good team relations or to avoid demotivating an employee. The AI-native approach helps the PM remove subjectivity, rely exclusively on facts from work processes, and prepare rock-solid arguments for growth-focused feedback before the 1-on-1 meeting. In this case, the AI does not argue with the manager's impressions; it checks the facts against the strict requirements of the policy.

Step 1. Collect artifacts (facts base for the PM)

The AI does not invent facts; it analyzes the provided data.
Before initializing the assessment, the PM must gather raw evidence:
The document "101 - Development Plan and Developer Assessment" (contains strict criteria for task levels and roles).
What scores the employee gave themselves and how they justify them.
Transcripts of work meetings (key artifact):
Daily meeting recordings (how the developer discusses tasks, whether they understand the business context).
Client meeting recordings (whether the camera is on, whether they speak the language of the business).
Technical sync recordings (whether synchronous code review is used, whether they help colleagues).
Project metrics: Lead Time (time from commit to production), release frequency, usage statistics for automated checks, and the presence or absence of test environments.

Step 2. Initialize the AI and load the context

Use advanced LLMs with a large context window (NotebookLM, Gemini 1.5 Pro, ChatGPT-4o).
Upload all collected text artifacts and send the AI the official system prompt approved by the policy:
You are evaluating a developer using the L10 / L20 / L30 / TECHLEAD / DEVOPS methodology.
Your task is to identify the real level, eliminate self-deception, curb overstatements, and define the next development epic. CORE PRINCIPLES:
There is no rating without evidence of independent delivery.
There is no L20 without full-stack value delivery.
There is no L30 without hypothesis-driven work and a new state.
If the client decomposed it, that is L10.
If the value was not used in production, it was not delivered.
Generic statements are not accepted - only facts.
Proceed step by step, asking me clarifying questions: STAGE 1.
Strict verification of each task against L10, L20, or L30. STAGE 3.
Assessment of the technical implementation (coupling check). STAGE 4.
TECHLEAD assessment (finding a real strategy for acceleration and team influence). STAGE 6.
Check for "engineering illusions." Stage 7.
Compare against the developer's self-assessment and identify overstatements. Stage 8.
Create an IDP in coaching mode (next development epic, 3 steps, metric, deadline).

Step 3. Go through the assessment stage by stage (dialog with AI)

Because the prompt includes a step-by-step algorithm, the AI will guide you through the process itself, acting as a strict mentor.
If the AI asks: "Was there direct contact with the user, and did the developer do both front and back independently?", answer honestly based on the transcripts.
Do not try to "push through" an inflated rating:
If you claim the task matches L30, the AI will challenge you with follow-up questions: "How did the project's meta-property change?
Were the hypotheses tested in production?".
If there is no evidence, the rating is automatically adjusted to L20 or L10 according to the rules.
Pass the AI the scores the developer assigned to themselves. The AI will identify discrepancies and prepare policy-based arguments for a constructive 1:1 conversation.

Step 4. Build the RS-DEVS report and development plan (PDP)

After successfully completing all 8 stages, ask the AI for the final summary.
You will receive a ready-made report to enter into the system:
Actual task level (L10 / L20 / L30).
Final scores for each RS-DEVS criterion with strict justification.
Feedback block: what exactly is the self-assessment overstatement, if any, and what is the main growth constraint. PDP-TXT (Next step for the development plan):
Define the next development epic, 3 concrete steps for the next 1-3 months, a success metric, and a deadline. With this document, the PM enters the 1-on-1 meeting with the developer, using company standards and indisputable facts, which shifts the conversation from personal preferences to transparent professional growth.

Bottom line: why AI-native assessment outperforms the traditional approach

The main idea of this approach is to separate facts from impressions.
When an algorithm based on policy stands between the manager and the developer, both sides benefit:
The developer gets a transparent, predictable growth system.
It knows exactly which facts are needed to move to the next level and does not depend on the manager's mood.
The manager gets rid of the discomfort of a "subjective verdict".
The AI forms the arguments from data - you only need to discuss them in a constructive dialogue.
The team sees the same rules for everyone: the same criteria, the same process, the same bar for all. AI does not replace the human conversation - it makes it fair. And fair feedback is the only kind of feedback that truly helps people grow.

Contacts

Let's Discuss Your Project

Leave your current contact details and describe your task. We will come back with clarifying questions and a proposal for the next step.