Build an AI coding agent with Docker, Python, and pytest

Published July 25, 2026

If you want to automate the boring parts of your dev workflow, building your own AI coding agent is a concrete way to do it. This walkthrough covers the full stack: Docker environment setup, Hugging Face model loading, automated testing with pytest, and Kubernetes deployment for production scale.

Table of Contents

️ What You Are Building

The agent uses a sequence-to-sequence model to generate code from natural language prompts. Automated test cases validate every output. The whole thing runs in Docker containers and scales via Kubernetes when you need more compute.

The core Python dependencies are:

transformers==4.28.0 (Hugging Face, provides pretrained models including GPT-3 and BERT variants)
numpy==1.23.5
pandas==1.5.3
scikit-learn==1.1.1

Step 1: Spin Up the Docker Environment

Pull a lightweight Python 3.11 image and start a named container with your working directory mounted:

docker pull python:3.11-slim
docker run -it --name ai-coding-agent -v $(pwd):/usr/src/app -w /usr/src/app python:3.11-slim bash

This binds your local directory to /usr/src/app inside the container so edits are reflected immediately. Keep a requirements.txt to lock dependency versions across environments.

Install dependencies inside the container:

pip install -r requirements.txt

3D rendered ai text on dark digital background

Step 2: Load the Model and Generate Code

The example uses microsoft/DialoGPT-medium as a starting point. It was designed for conversational tasks but can be adapted for code generation:

from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

model_name = 'microsoft/DialoGPT-medium'

tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(model_name)

inputs = tokenizer("def greet(name):", return_tensors="pt")
outputs = model.generate(inputs["input_ids"])
print(tokenizer.decode(outputs[0]))

The tokenizer converts your code string into input tensors. The generate call produces a completion. Pretrained models let you skip training from scratch, but you will want to fine-tune on domain-specific code using transfer learning for better accuracy.

Step 3: Automate Testing with pytest

Generated code is only useful if it actually runs correctly. Wire pytest into your pipeline to validate every output automatically:

import pytest

def add_numbers(a, b):
    return a + b

def test_add_numbers():
    assert add_numbers(1, 2) == 3
    assert add_numbers(-1, 1) == 0
    assert add_numbers(-1, -1) == -2

# pytest test_script.py

Run this inside your CI/CD pipeline on every commit. Defective code generation surfaces immediately rather than after a deploy.

a computer circuit board with a brain on it

Step 4: Containerize and Deploy

Once the agent works locally, package it into a Docker image for consistent deployment:

FROM python:3.9
WORKDIR /usr/src/app
COPY . .
RUN pip install --no-cache-dir -r requirements.txt
CMD ["python", "app.py"]

Deploy the image to Kubernetes for scaling and resource management. Kubernetes handles auto-scaling so you are not manually provisioning instances when compute demand spikes.

⚠️ Common Deployment Pitfalls

Resource exhaustion: AI models are CPU and GPU hungry. Allocate enough memory before you deploy to production.
Network latency: Real-time applications are sensitive to it. Use network-optimized instances and consider geographic availability zones.
Scaling misconfiguration: Use Kubernetes auto-scaling rather than fixed replica counts to avoid either starving or over-provisioning.
Security gaps: Harden your containers and use Kubernetes RBAC to control access to the agent.

Pro Tips for Production

Apply model quantization to shrink neural network size and cut inference compute costs.
Add caching to avoid repeating identical computations on common prompts.
Implement feedback loops so the model learns from test failures over time.
Include performance regression tests in your CI pipeline to catch degradation before it hits users.
Monitor with Grafana and Prometheus to track system health continuously.

Cagri Sarigoz( Founder )

I’m founder of BizStack at Cagri Sarigoz LLC and a passionate advocate for entrepreneurs.

With over 14 years in tech, marketing, and AI, including my role as Head of SEO at CitizenShipper and co-founder of TaleBot at Intale AI, I’m dedicated to sharing genuine, useful product insights and tips.

At BizStack, I aim to cut through the digital noise to provide clear, actionable advice.

And more than all else, I’m a father to a (always) little girl and a husband.

Contact me at [email protected] for assistance.