•
Ten verified 2026 coding-agent tools for after-work Python and first data/quant steps. Not a ranking. You still run a real .py file.
You do not need an Agentic AI bootcamp to stay useful at work. You need a 30–90 minute sitting, a real .py file, and a tool that will edit or explain code without taking over your job. This list is for employed people studying Python, data, and first quant steps after work — the QuantInsti/CFI plus Le Wagon evening pattern, not a career-switch camp and not a get-rich pitch.
The market is already selling “agentic” credentials. That is a different product. See The 2026 quant-education market is pitching Agentic AI as the enrollment hook. This page is tools you can open tonight.
If Python is not on the machine yet, start with Set up Python after work in 30 minutes (Windows, Mac, VS Code). For pacing, Learn Python after work: 19 short lessons, a few evenings a week. First useful script: Your first Python program that uses a library (not another print).
Not a ranking from search volume or CTR. Checked 1 September 2026 against official pages. A tool had to still ship, help in one after-work sitting, and touch Python, data, or first quant steps.
Dropped: Continue (acquired by Cursor; standalone product wound down — continue.dev). Windsurf as a brand (now Devin Desktop). ChatGPT agent mode (OpenAI help: no longer available; use ChatGPT Work). No Midjourney. No Notion AI.
What it is. An AI-native code editor whose headline is the coding agent — local Agent plus cloud agents.
After-work sitting (45–75 min). Open last night’s folder. Ask the agent to add pandas.read_csv and print(df.shape). You run python that_file.py. Keep the traceback if it fails.
Sourced fact (2026). Cursor pricing: Hobby is free with limited Agent requests; Individual Pro is $20/month, with extended Agent limits and cloud agents.
Caveat. The agent will invent pandas methods. The diff is a draft. You still run the file.
What it is. GitHub’s pair programmer: completions, chat, agent mode in the editor, a cloud agent that can open a pull request, and Copilot CLI.
After-work sitting (30–60 min). Stay in VS Code. Agent mode: filter one CSV column and print the count. Review the diff. Run the file.
Sourced fact (2026). Copilot plans: Pro is $10/month. Chat, agent mode, code review, cloud agent, and CLI consume GitHub AI Credits (1 credit = $0.01). Completions stay unlimited on paid plans. Free supports CLI and agent mode with limited usage. Credits went live 1 June 2026 (changelog).
Caveat. Unlimited Tab is not unlimited agents. A long agent session can burn the month’s credits.
What it is. Anthropic’s agentic coding tool: reads the repo, edits files, runs commands. Surfaces: terminal, IDE, desktop, browser.
After-work sitting (45–90 min). In the project folder: claude, then “add a groupby to this CSV script, run it, fix the traceback.” Watch the commands. Keep the .py.
Sourced fact (2026). Claude pricing: Claude Code is included in all paid plans. Pro is $20/month monthly, or $17/month annual ($200 up front). Free has chat; Claude Code is on paid plans. Terminal and chat share one pool.
Caveat. A heavy coding evening can hit the rolling five-hour window. Do not start three parallel agents if you still need chat for work email.
What it is. OpenAI’s coding agent (CLI, IDE extension, ChatGPT, desktop app) — Codex in ChatGPT. Not the 2021 research model recycled as a blog title.
After-work sitting (40–80 min). Open Codex on the local folder. “Write a matplotlib line chart from this CSV and save chart.png.” You run the script. You open the PNG.
Sourced fact (2026). Codex and ChatGPT plans: Codex is included across plans, including Free and Go; limits vary. Plus lists expanded Codex usage on ChatGPT pricing — check that page for current USD.
Caveat. Codex, ChatGPT Work, and Workspace Agents share an allowance. One long cloud task can spend a week of after-work sessions.
What it is. ChatGPT’s longer-task surface, plus existing custom GPTs. Agent mode is no longer available. New GPT creation is closed on personal Free, Go, Plus, and Pro accounts; existing GPTs remain usable.
After-work sitting (30–50 min). Paste a traceback. Ask Work to explain the line, not to invent a trading bot. Apply the fix in the .py. Use an existing GPT if you have one; do not plan a new personal GPT.
Sourced fact (2026). Plus includes Work on desktop, web, and mobile on ChatGPT pricing. Pro tiers: $100/month is 5× Plus; $200/month is 20×. Check the pricing page for Plus/Go USD.
Caveat. This is not an editor. Copy-paste is the loop. A chat answer is not a backtest.
What it is. Google’s agent-first IDE, CLI, and SDK (antigravity.google). This is the 2026 Gemini coding-agent surface; Gemini CLI was retired in favour of Antigravity.
After-work sitting (45–75 min). One workspace, one agent: plot a single series and write a test that the PNG exists. Run it locally.
Sourced fact (2026). Antigravity pricing: Individuals $0/month with basic weekly rate limits and unlimited Tab. Google AI Pro and Ultra raise limits; paid USD is bundled with Google AI — check that page.
Caveat. Free weekly limits are real. Do not plan a 90-minute agent marathon on Individual without checking quota.
What it is. Cognition’s local IDE and Agent Command Center. Windsurf is now Devin Desktop (June 2026 over-the-air). Plans carried over.
After-work sitting (45–75 min). If you used Windsurf, you already have this. Point a local agent at one pandas groupby. Review the diff. Run the file.
Sourced fact (2026). Devin pricing: Free $0 (light quota, unlimited Tab); Pro $20/month includes Desktop; cloud agents are listed on Pro.
Caveat. Docs that still say Cascade or Windsurf are stale. Confirm Devin Desktop before you follow a 2025 install gist.
What it is. Open-source terminal pair programming against your git repo (aider.chat). Maps the codebase, edits, auto-commits. You bring the LLM.
After-work sitting (30–60 min). cd into a small repo. Ask for one function. Read the commit. Run tests. git show if you dislike the message.
Sourced fact (2026). The software is free (python -m pip install aider-install). You pay the model provider, or run a local model.
Caveat. The homepage still names older models (Claude 3.7, o3-mini). Pick a current model in the docs. Auto-commits surprise people who expected a dirty working tree.
What it is. Open-source autonomous agent in the editor (and a CLI). Plan then Act. Edits files, runs commands, asks permission by default. VS Code id: saoudrizwan.claude-dev (Marketplace).
After-work sitting (40–70 min). Plan mode: “clean this CSV, write clean.py.” Act. Approve each step. Run python clean.py.
Sourced fact (2026). cline.bot lists it as free/open source (Apache 2.0) with bring-your-own key or a Cline account. You pay the model.
Caveat. Auto-approve is optional. Leave approval on until the task is boring. Check the Marketplace publisher — forks exist.
What it is. Cognition’s autonomous engineer on a remote machine — distinct from the local Desktop IDE. cognition.com, sign-up at app.devin.ai.
After-work sitting (kick off 15 min, review 30 min later). One bounded task: “add a README and a pytest for clean.py.” Do not ask it to “build a strategy.” Read the diff.
Sourced fact (2026). Devin pricing: Pro $20/month lists cloud agents (Devin Cloud). Free is light quota without the same cloud access. Teams starts at $80/month minimum.
Caveat. Overkill for a 45-minute pandas sitting. Useful when you can write a tight spec and walk away. You still review the code.
.py fileAgents draft. They do not replace the loop you are paying for: write, run, read the traceback, change one thing. That is how you stay competitive in the job you have.
If you want that loop as short lessons instead of a three-hour classroom block, Introduction to Python Programming is 19 published lessons. Check the course page for enrollment. It is not a bootcamp, not a subsidised certificate, and not a quant-trading credential.