Week 45, updated 2 Nov 2026

AI use cases Automation & agent frameworks Build AI agents in Python or TypeScript

Build AI agents in Python or TypeScript

Pick a code-first agent framework for Python or TypeScript: tools, structured output, evaluation and deployment. Workflow and ranked frameworks.

Tools compared
6
Open source
6of 6
Reviewed
16 Oct

For: Developers

Best-scored tools for this jobscore, safety
  1. 1LangChainAgent framework54Caution
  2. 2Pydantic AIAgent framework46Caution
  3. 3MastraAgent framework44No issues found
  4. 4Agent Development Kit (ADK, Python)Agent framework43Caution
  5. 5OpenAI Agents SDK (Python)Agent framework39Caution
  6. 6smolagentsAgent framework30No dependencies scanned
What the score and shapes mean

Score = public maintenance, adoption and docs signals, as a percentile of 52 tools. Nobody here tested these hands-on. Shapes are the separate safety label: No issues found, No dependencies scanned, Caution.

What it does for you

Developers adding agent features to a product or building internal automation.

  • Give a model tools (functions, APIs) and let it decide when to call them
  • Return validated, typed output
  • Run multi-step agents with state and retries
  • Trace and evaluate agent runs
  • Deploy an agent as a service
More about this job

Agent frameworks provide the loop that calls a model, runs tools and returns structured results. The differences that matter are language, how much they abstract, typing, observability and how they handle evaluation and deployment.

Workflow in 5 steps

  1. Write the task as a plain function first.
  2. Choose a framework that matches your language and that you can read the source of.
  3. Start with one tool and typed output.
  4. Create a small evaluation set of realistic inputs and expected behaviour.
  5. Add tracing and cost limits before deploying.
Full steps
  1. Write the task as a plain function first; decide what truly needs a model.
  2. Choose a framework that matches your language and that you can read the source of.
  3. Start with one tool and typed output; add tools one at a time.
  4. Create a small evaluation set of realistic inputs and expected behaviour.
  5. Add tracing and cost limits before deploying.

Prompts to copy

Starting points we wrote, not tested results.

Write a minimal agent with one tool, get_weather(city), that returns typed output. Include the code to run it and one test with a stubbed tool.

2 more prompts

Add a retry and a timeout around the tool call, and log each step with its inputs and outputs.

List the failure modes of this agent (bad tool arguments, empty results, loops) and add a test for each.

Watch out for

  • Agent loops can burn through paid API tokens quickly.
  • Agents that can run commands can delete or overwrite files.
  • AI output can be fluent and wrong.
  • Check each tool's licence before commercial use, and check the licence of any code or text the AI reproduces.
  • Frameworks change fast.
  • Licence to check before commercial use: Mastra (Apache-2.0 (ee/ excluded)).
  • CautionLangChain7 critical-severity known vulnerabilities in runtime dependencies (OSV)
  • CautionPydantic AI1 critical-severity known vulnerability in runtime dependencies (OSV)
  • No issues foundMastraNo issues found in this week’s scan
  • CautionAgent Development Kit (ADK, Python)2 critical-severity known vulnerabilities in runtime dependencies (OSV)
  • CautionOpenAI Agents SDK (Python)3 critical-severity known vulnerabilities in runtime dependencies (OSV)
  • No dependencies scannedsmolagentsCode checks clean, no lockfile to check
Why these warnings?
  • Agent loops can burn through paid API tokens quickly. Set spending limits with your model provider and watch the first few runs.
  • Agents that can run commands can delete or overwrite files. Work in a branch or sandbox, commit first, and approve destructive commands by hand.
  • AI output can be fluent and wrong. Check facts, code and citations yourself before you rely on them or publish them.
  • Check each tool's licence before commercial use, and check the licence of any code or text the AI reproduces.
  • Frameworks change fast. Pin versions and check release cadence and open issues before committing.
  • LangChain: 7 critical-severity known vulnerabilities in runtime dependencies (OSV)
  • Pydantic AI: 1 critical-severity known vulnerability in runtime dependencies (OSV)
  • Agent Development Kit (ADK, Python): 2 critical-severity known vulnerabilities in runtime dependencies (OSV); Skill/agent file instructs piping a remote script into a shell: .agents/skills/adk-setup/SKILL.md — curl -LsSf https://astral.sh/uv/install.sh | sh
  • OpenAI Agents SDK (Python): 3 critical-severity known vulnerabilities in runtime dependencies (OSV)
  • smolagents: code and instruction-file checks found nothing, but there is no runtime lockfile or manifest, so dependencies were not checked

Recommended tools

From our ranked data, highest score first. A high score does not mean a clean scan.

Related jobs

Sources for tool facts

langchain-ai/langchain · pydantic/pydantic-ai · mastra-ai/mastra · google/adk-python · openai/openai-agents-python · huggingface/smolagents. Data fetched 2 Nov 2026. Scores and labels are automated, relative and not a hands-on test.

Get the board every Monday

One email a week: the new top 10, the biggest score changes and every safety label that changed.

Free. Double opt-in: you confirm by email first, unsubscribe in one click.