Learning and Python fundamentals
Python learning paths, core language concepts, algorithms, computer-science practice, and professional skill development.
36%
Best tweets about Python
50 Best Tweets About Python (2026)
Explore the best tweets about Python, including language features, libraries, data work, AI development, automation, performance, and engineering practices.
Technical Python code, libraries, tooling, releases, performance, automation, data science, AI engineering, and production lessons.
- Creators
- 40
- Updated
Original Xholic analysis
What 50 top Python posts reveal
Across 50 Python posts, learning and fundamentals was the largest identified theme, while developer tooling, agentic coding, AI/ML engineering, performance, and production practices were also recurrent. Deterministic analytics show higher median all-time scores for media posts than text-only posts, and the largest score outliers included a compact GPT implementation, an AgentScope announcement, and a random.seed() analysis.
- Dominant tone
- Positive
- Median score
- 12.5
- Leading format
- Other
- Recent posts
- 36%
84% of posts
All-time engagement
100% of posts
Published in 90 days
Conversation map
The themes creators return to
Developer tooling and packaging
Python tooling for environments, linting, formatting, typing, packaging, CLIs, IDEs, and developer workflows.
26%
AI agents and agentic coding
Python-based AI agents, coding agents, MCP integrations, persistent runtimes, and agent-oriented frameworks.
24%
AI and ML engineering
Python libraries and workflows for LLM applications, RAG, document ingestion, AI APIs, and model training or inference.
22%
Performance and runtime internals
Profiling, debugging, memory inspection, GPU acceleration, JIT compilation, and Python runtime performance.
22%
Backend and production engineering
Backend and production engineering with APIs, frameworks, databases, authentication, testing, deployment, and maintainable software practices.
20%
Open-source ecosystem and projects
Open-source Python packages, ecosystem projects, maintainers, licenses, and community or institutional support.
14%
Data engineering and analytics
Data engineering and analytics with files, ETL, DataFrames, warehouses, validation, visualization, and time-series workflows.
8%
Tone and stance
Sentiment Positive leads
Author posture Supportive leads
Performance benchmark
- Median likes
- 42
- Median reposts
- 7
- Median replies
- 3
- Median views
- 2.7K
Posts with media make up 58% of this collection. Their median all-time score is 21.4, compared with 3.75 for text-only posts.
Format mix
- Other 100% · score 12.5
Consensus and debate
Where creators agree, and where they do not
Shared view
Fundamentals are connected to practical engineering work
Several learning and roadmap posts connect core Python concepts—such as data structures, functions, packaging, APIs, databases, automation, testing, profiling, deployment, and projects—to practical engineering work.
Shared view
Agent work is a recurring Python topic
Agent-focused posts describe Python frameworks and designs involving MCP support, sandboxed code execution, event histories, persistent state, and a separation between model-driven behavior and deterministic Python code.
Shared view
Profiling and debugging are emphasized before performance work
The performance and debugging posts focus on identifying slow code with profiling and preserving or inspecting state when long-running Python processes fail or hang.
Open debate
Python productivity versus production-AI concerns
One post recommends choosing the language that maximizes team productivity when performance is not needed; another argues that Python is poorly suited to production AI infrastructure. These posts present competing views without defining a shared production boundary.
Open debate
AI assistance still requires code understanding
Learning-oriented posts argue that developers should understand, review, and guide AI-generated code. Another post attributes variation in AI coding quality across languages to the availability of public training examples.
What performs
Patterns behind standout posts
A compact GPT implementation and agent-framework announcement led outliers
The overall median all-time score was 12.54. The three highest-score outliers were the dependency-free GPT implementation (6129.63), the AgentScope announcement (2133.47), and the random.seed() sign-collision analysis (702.18).
Media posts had the higher median score
Media appeared in 29 of 50 posts (58%). Its median all-time score was 21.374, compared with 3.752 for text-only posts.
Fundamentals led theme volume; data engineering led theme median
Learning and Python fundamentals was the largest theme at 36% (18 tweets). AI and ML engineering had a median all-time score of 45.357, while data engineering and analytics had the highest theme median at 48.76.
Statistical standouts
- View standout post 1 Score 6129.6 · 488.81× median
- View standout post 2 Score 2133.5 · 170.13× median
- View standout post 3 Score 702.2 · 56× median
- View standout post 4 Score 589.7 · 47.03× median
- View standout post 5 Score 414.7 · 33.07× median
Creator landscape
Who shapes this conversation
The five most represented creators account for 20% of the selected posts.
-
1. Vaishnavi
@_vmlops
2 posts
-
2. Shalini Goyal
@goyalshaliniuk
2 posts
-
3. Solomon Eseme
@Kaperskyguru
2 posts
-
4. Andrej Karpathy
@karpathy
2 posts
-
5. Kirk Borne
@KirkDBorne
2 posts
-
6. Mahesh
@MaheshPawaar
2 posts
Karpathy combines compact implementation with a runtime footgun
Andrej Karpathy shared a 243-line, dependency-free Python GPT implementation and separately documented a random.seed() sign-collision pitfall, including its implications for train/test splits.
Vaishnavi highlights Python-library capabilities
Vaishnavi highlighted Manim for mathematical visualization and noted OpenAI Python-library features including retries, streaming, pagination, async support, workload-identity authentication, and webhook verification.
Goyal organizes Python around applied pathways
Shalini Goyal published roadmap-style posts connecting Python concepts and libraries to data-engineering workflows and quantum-computing use cases.
How this analysis was made
Themes, sentiment, stance, and post format are classified per tweet. All counts, shares, medians, creator concentration, freshness, and performance comparisons are then calculated directly from the published snapshot.
Xholic's all-time score compares engagement while accounting for reach, post age, and creator consistency. It is used for relative comparisons within this collection.
This report analyzes the exact 50-post snapshot shown below. AI identifies editorial categories and drafts explanations; all statistics are calculated from the snapshot, and every narrative claim is checked against cited posts before publication.
Best Python tweets
Top Python tweets from 40 creators
Ranked 01–50
- 01
- 02
@hasantoxr ·
🚨 BREAKING: CHINA just released a Python framework for building AI agents. 100% OPEN SOURCE. It has visual agent design, MCP tools, memory, RAG, and reasoning. All built in. All working together. It's called AgentScope. You describe your agent system. It builds the architecture, wires the tools, and runs the whole thing. You come back and there's a working multi-agent pipeline. Not a prototype. Not a demo. The actual system. Not a wrapper. Not a chatbot builder. A full Agent-Oriented Programming framework that thinks in agents from the ground up. Here's what it does out of the box: → Visual agent builder so you design your entire system before writing a single line of code → Native MCP tool support, plug any external tool directly into any agent in your pipeline → Built-in memory so every agent remembers context, decisions, and history across sessions → RAG pipeline ready to connect your own documents, databases, and knowledge bases → Reasoning modules that let agents plan, reflect, and self-correct without human input → Multi-agent coordination so your agents collaborate as a system, not a pile of isolated API calls Here's how it thinks: You define your goal. AgentScope maps the agent roles. Each agent gets its tools, its memory, its reasoning layer. They coordinate. Results flow back up. You get a finished output. A single complex task might route through a planner agent, a researcher agent, a coder agent, and a critic agent, each doing its job, then converge into one clean deliverable. Here's the wildest part: AgentScope is built by Alibaba DAMO Academy. The same lab behind Qwen. They didn't assemble this from existing pieces. They designed the entire framework from first principles around how agents actually need to think, remember, and work together. Most frameworks give you building blocks. AgentScope gives you an architecture. The community has already started plugging it into data pipelines, research workflows, and full automation systems the team never planned for. 100% Open Source. Apache 2.0 License.
- 03
@karpathy ·
In today's episode of programming horror... In the Python docs of random.seed() def, we're told "If a is an int, it is used directly." [1] But if you seed with 3 or -3, you actually get the exact same rng object, producing the same streams. (TIL). In nanochat I was using the sign as a (what I thought was) clever way to get different rng sequences for train/test splits. Hence gnarly bug because now train=test. I found the CPython code responsible in cpython/Modules/_randommodule.c [2], where on line 321 we see in a comment: "This algorithm relies on the number being unsigned. So: if the arg is a PyLong, use its absolute value." followed by n = PyNumber_Absolute(arg); which explicitly calls abs() on your seed to make it positive, discarding the sign bit. But this comment is actually wrong/misleading too. Under the hood, Python calls the Mersenne Twister MT19937 algorithm, which in the general case has 19937 (non-zero) bits state. Python takes your int (or other objects) and "spreads out" that information across these bits. In principle, the sign bit could have been used to augment the state bits. There is nothing about the algorithm that "relies on the number being unsigned". A decision was made to not incorporate the sign bit (which imo was a mistake). One trivial example could have been to map n -> 2*abs(n) + int(n < 0). Finally this leads us to the contract of Python's random, which is also not fully spelled out in the docs. The contract that is mentioned is that: same seed => same sequence. But no guarantee is made that different seeds produce different sequences. So in principle, Python makes no promises that e.g. seed(5) and seed(6) are different rng streams. (Though this quite commonly implicitly assumed in many applications.) Indeed, we see that seed(5) and seed(-5) are identical streams. And you should probably not use them to separate your train/test behaviors in machine learning. One of the more amusing programming horror footguns I've encountered recently. We'll see you in the next episode. [1] https://t.co/srv1ZBlDsi [2] https://t.co/qpnKdvfVNS
- 04
- 05
@_vmlops ·
Grant Sanderson BUILT THE TOOL THAT MAKES MATH LOOK LIKE ART it's called Manim built by the mind behind 3Blue1Brown every smooth, beautiful math visual you've seen there..? written in code here's what actually happened though he wasn't trying to change math education he was just annoyed the tools were bad...the visuals were ugly... the math deserved better so he did what engineers do when nothing fits he built it himself python, opengl...late nights...no audience just a guy who had a picture in his head & needed a way to put it on screen then the internet found him the backprop, the attention mechanism, the transformers, the gradients the entire mathematics the AI world runs on he made it look simple using a tool he built in his room ml engineers, math students, confused undergrads at 2am all looking for the same thing someone to make the hard stuff feel like human & Grant did exactly that not by simplifying the math but by making you see it that's the thing about the best tools they don't come from product teams they come from someone who couldn't find what they needed `pip install manimgl` → https://t.co/6NVQxCMjRn
- 06
@avrldotdev ·
Step-1: Learn Python Step-2: Understand data structures, iterators & generators Step-3: Master memory management, GIL & concurrency (threadingmultiprocessing/asyncio) Step-4: Build a CLI tool using argparse & package it with PIP Step-5: Develop REST APIs with FastAPI/Django Step-6: Work with DBs & ORMs Step-7: Build data pipelines & automation scripts that scale Step-8: Optimize performance using profiling tools (cProfile/line_profiler) Step-9: Deploy using Docker & K8s Step-10: Ship it
- 07
@GithubProjects ·
Pyinstrument is a Python profiler that helps you identify the slowest parts of your code so you can focus optimization efforts. - Supports Python 3.8+ and installs via pip. - Renders interactive HTML profiles with timeline mode. - Integrates with Django middleware and FastAPI. - Adapts precision of printed durations to profiling interval.
- 08
@nrqa__ ·
🚨 BREAKING: Someone at Microsoft just open-sourced a tool that converts almost any file format into clean Markdown. It's called MarkItDown. And it handles everything. PDFs. Word docs. PowerPoints. Excel spreadsheets. Images. Audio files. HTML. ZIP archives. One tool. One output format. Clean Markdown every time. Here is why this matters right now: Every AI coding agent, every LLM, every RAG pipeline works better with Markdown than with raw file formats. PDFs confuse them. DOCX files bloat the context. PowerPoints are basically unreadable. MarkItDown fixes the input layer. Drop any file in. Get clean, structured Markdown out. Feed it directly to your AI agent. Full format support: -> PDF documents -> Word files (DOCX) -> PowerPoint presentations (PPTX) -> Excel spreadsheets (XLSX) -> Images with EXIF metadata and OCR -> Audio files with speech transcription -> HTML pages -> ZIP archives (processes contents recursively) Use cases that are immediately obvious: -> Feed a PDF research paper to Claude or GPT without fighting the format -> Convert a client PowerPoint into structured notes in seconds -> Preprocess entire document folders for RAG pipelines -> Build document ingestion pipelines with zero formatting headaches Install in one line: pip install markitdown That is it. Works as a Python library or a CLI tool. From Microsoft. Actively maintained. Already one of the most starred Python tools on GitHub. 100% Open Source. MIT License.
- 09
@johncrickett ·
BitTorrent wasn't built with Python because it was fast. But because it didn't matter. At its peak it carried a third of all internet traffic. Bram Cohen could have built it in anything. He picked Python, saying: "People who are into Python aren't actually into Python. They just want to get the work done." Unless you need performance, pick the language that makes you and your team productive.
- 10
@aiwithjainam ·
10 GITHUB REPOS THAT BUILD A FULL TRADING AND FINANCE STACK. Bookmark every single one. Hedge funds pay six figures for less than this. 1. Fincept Terminal - https://t.co/JV8Te5qWZ9 Open source Bloomberg Terminal. CFA Level 1-3 analytics, 20+ investor AI agents, 100+ data connectors, and a 3D globe tracking live ships and aircraft. 2. Vibe-Trading - https://t.co/W3YMOLYbue Multi-agent AI trading system with 64 finance skills, 29 hedge fund presets, cross-market backtesting, and full quant tools. 3. AutoHedge - https://t.co/cNZ4y7nvIm Autonomous hedge fund in Python. 4 AI agents handle strategy, risk, sizing, and execution. Live on Solana right now. 4. OpenBB Terminal - https://t.co/fZ4AXhfnY2 Investment research platform with stocks, crypto, forex, options, economy, and alternative data in one interface. 5. Qlib - https://t.co/8jn9ts4qmI Microsoft's AI-oriented quant investment platform. Supports the entire ML pipeline from data to portfolio optimization. 6. FinGPT - https://t.co/lvpe1hcyV2 Open source financial LLMs. Sentiment analysis, market forecasting, and trading signals trained on financial data. 7. Freqtrade - https://t.co/Ze3B41IHlm Free crypto trading bot written in Python. Backtesting, machine learning optimization, and Telegram control built in. 8. Backtrader - https://t.co/Odwbyp4Ayd The most popular Python backtesting framework. Test any strategy on any asset with live trading support. 9. Lean - https://t.co/ddOAUFEgIS QuantConnect's algorithmic trading engine. Used by hedge funds and runs on C#, Python, and F#. 10. FinanceToolkit - https://t.co/KKjSBdj87W 200+ financial ratios, indicators, and performance measurements in one Python library. Replaces Bloomberg formulas. Wall Street built a moat with software. Open source just drained it.
- 11
@techNmak ·
Someone quietly built a computer science degree inside a GitHub repository. Sorting. Graphs. Dynamic programming. Data structures. Cryptography. Machine learning. All implemented in Python. Then you see the folders: maths sorts graphs hashes matrix ciphers geodesy physics quantum strings fractals geometry graphics knapsack searches financial blockchain scheduling conversions electronics fuzzy_logic backtracking audio_filters file_transfer project_euler greedy_methods linear_algebra neural_network boolean_algebra computer_vision data_structures networking_flow web_programming bit_manipulation data_compression machine_learning cellular_automata genetic_algorithm divide_and_conquer linear_programming dynamic_programming digital_image_processing Dive in, explore the topics that interest you, and learn something new. GitHub Repo in comments.
- 12
@goyalshaliniuk ·
Python becomes much easier to learn when you connect it directly with real data engineering work. For data engineers, Python is not just a programming language. It helps you read files, clean messy datasets, connect APIs, move data between systems, automate pipelines, query databases, and prepare workflows for production. The journey starts with the core foundations: Variables, data types, loops, functions, modules, packages, error handling, and virtual environments. Then it moves into data structures and logic, where lists, dictionaries, tuples, sets, iterators, generators, and comprehensions help you organize and process data better. After that, Python becomes practical. You learn how to work with CSV, JSON, Excel, XML, Parquet, Avro, ORC, and file paths. You use libraries like Pandas, NumPy, Polars, PyArrow, Dask, OpenPyXL, and FastParquet to clean, transform, validate, and prepare datasets. Then come the real data engineering layers: ETL workflows. APIs and external data sources. Databases and warehouses. Pipeline orchestration. Big data and cloud engineering. Production-ready projects. This is where Python starts feeling like a complete data engineering toolkit. The goal is not to learn every library at once. The goal is to understand where each concept fits in the pipeline. Learn the basics. Practice with files. Build ETL workflows. Connect APIs. Work with databases. Automate pipelines. Ship real projects. That is how Python turns from syntax into real data engineering skill.
Watch video
- 13
@PythonDvz ·
🤔 Feeling lost in Python? Save this roadmap and follow it step by step from beginner to advanced without wasting time 📌 Your Python roadmap for 2026 is finally here 🐍🔥 📍Month 1: Python fundamentals Syntax, variables, data types, operators, conditions, loops, functions, and type casting 📍Month 2: Data structures Lists, tuples, sets, dictionaries, and strings 📍Month 3: OOP Classes, objects, inheritance, methods, abstract classes, dunder methods, and super() 📍Month 4: Handling Exceptions, try except, finally, and file handling 📍Month 5: Modules Built in modules, requests, numpy, pandas, and custom modules 📍Month 6: Advanced Python Map, filter, zip, enumerate, lambda, decorators, generators, regex, and pip Master these in order and your Python journey will become much clearer 🚀 #python #pythonroadmap #learnpython #programming #coding
- 14
@heynavtoor ·
This is the tool AWS does not want you to install. Amazon Rekognition charges about $0.001 per image to detect objects. Running that on a busy video feed hits a million images fast. A million images is $1,000 in AWS fees. Google Vision AI charges $1.50 per 1,000 images. Azure Computer Vision bills per transaction. Roboflow's own paid plan runs $99 to $999 a month. There is a free Python library that does the same work on your laptop. It is called Supervision. pip install supervision Here is the story. A guy in Kraków got hired to write marketing demos for a computer vision startup. His job was making little Python scripts for Twitter that showed off what AI models could do. Traffic counting. Store analytics. Sports tracking. Every demo needed the same boilerplate. Draw a box around the car. Draw a line across the road. Count the crossings. Track each car with a stable ID. Save the output as a video. He got tired of rewriting the same code every week. So he pulled the shared code into a Python package. He put it on GitHub under an MIT license. He shipped it. That was November 2022. Today: 48,212 stars. 4,423 forks. 1.2 million downloads every month on PyPI. 160+ contributors. Powers computer vision pipelines at real companies. Point a webcam at a road. Plug in a free object-detection model like YOLO. Supervision draws boxes on every car. Tracks each one. Counts what crosses your line. Estimates speed. In real time. That is a traffic monitoring system. Point the camera at a store entrance. It counts every shopper who walks in. Tracks their path. Builds a heatmap of foot traffic. That is retail analytics. Companies pay tens of thousands a year for the same output. Point the camera at a basketball court. It tracks every player, IDs their jersey number with a second model, logs their positions frame by frame. The guy in Kraków built the basketball keypoint model for this over a weekend and posted it on LinkedIn. His name is Piotr Skalski. His GitHub bio says: "I open-source stuff." Amazon can't take this back. The license does not permit that. Google can't take this back. The license does not permit that. Roboflow could have kept it internal. They open sourced it instead. Your camera already sees everything. Piotr Skalski wrote the library that lets it understand what it sees. For free. (Link in the comments)
- 15
@goyalshaliniuk ·
Quantum computing is moving beyond research labs, and Python is becoming one of the easiest ways to explore it. From circuit simulation to quantum machine learning, optimization, chemistry, hardware control, and error mitigation, the ecosystem now offers a library for almost every use case. Here are 15 Python libraries worth knowing: → 𝗤𝗶𝘀𝗸𝗶𝘁 — Build, optimize, simulate, and run gate-based circuits. → 𝗖𝗶𝗿𝗾 — Design hardware-aware circuits for noisy processors. → 𝗣𝗲𝗻𝗻𝘆𝗟𝗮𝗻𝗲 — Create differentiable programs and hybrid ML models. → 𝗧𝗲𝗻𝘀𝗼𝗿𝗙𝗹𝗼𝘄 𝗤𝘂𝗮𝗻𝘁𝘂𝗺 — Combine Cirq with TensorFlow for quantum AI. → 𝗢𝗽𝗲𝗻𝗙𝗲𝗿𝗺𝗶𝗼𝗻 — Model molecules, fermionic systems, and quantum chemistry. → 𝗔𝗺𝗮𝘇𝗼𝗻 𝗕𝗿𝗮𝗸𝗲𝘁 𝗦𝗗𝗞 — Run jobs on simulators and quantum hardware. → 𝗤𝗶𝗯𝗼 — Support simulation, calibration, and hardware control. → 𝗤𝘂𝗹𝗮𝗰𝘀 — Run fast simulations of large quantum circuits. → 𝗖𝗨𝗗𝗔-𝗤 — Build hybrid workflows across CPUs, GPUs, and QPUs. → 𝗤𝘂𝗧𝗶𝗣 — Simulate states, operators, dynamics, and open systems. → 𝗽𝘆𝘁𝗸𝗲𝘁 — Compile and optimize circuits across hardware platforms. → 𝗗-𝗪𝗮𝘃𝗲 𝗢𝗰𝗲𝗮𝗻 — Solve optimization problems with quantum annealing. → 𝗽𝘆𝗤𝘂𝗶𝗹 — Build and execute programs for Rigetti systems. → 𝗠𝗶𝘁𝗶𝗾 — Reduce noise using error-mitigation techniques. → 𝗠𝗶𝗰𝗿𝗼𝘀𝗼𝗳𝘁 𝗤𝗗𝗞 — Combine Python and Q# for quantum development. Start with Qiskit or Cirq, then choose based on your goal: AI, chemistry, optimization, simulation, or hardware flexibility. Which quantum Python library would you explore first?
Watch video
- 16
@sukh_saroy ·
🚨Breaking A Python library that reverse-engineers Google Flights' internal API just dropped -- and it connects directly to Claude as an MCP server. It's called fli. And it's not a wrapper around a flight search UI. It hits Google's internal endpoints directly -- no HTML parsing, no browser automation, no Puppeteer -- and returns structured flight data fast. Here's what it can do: → Search one-way and round-trip flights with departure date and cabin class → Filter by departure time window, specific airlines, max stops, and price ceiling → Sort results by price, duration, departure time, or arrival time → Find cheapest dates across any date range with a sparkline price chart → Run as an MCP server so Claude can search flights from natural language → Built-in rate limiting, retry logic with exponential backoff, and input validation Here's the wildest part: Google Flights' internal API doesn't require traditional authentication. fli discovered that, reverse-engineered the encoding, and packaged the whole thing into a clean Python library with Pydantic models and a CLI. Ask Claude "what's the cheapest non-stop flight from JFK to LHR next month in business class?" and it actually answers with real Google Flights data. One command to install: `pipx install flights` One command to wire up Claude Desktop: `fli-mcp` 100% Open Source. MIT License. (Link in the comments)
- 17
@Shruti_0810 ·
Don’t overthink Python. Start with this: • Python Basics → syntax & control flow • Data Structures → lists → dicts → sets → tuples • Functions → reusable logic • OOP → structure bigger programs • Asyncio → concurrency • FastAPI / Flask → build APIs • Testing → reliable code • Packaging → reusable libraries That’s enough to build real systems. You don’t need 500 Python libraries. You need strong fundamentals. 🐍
- 18
@smratitiwa86867 ·
They tried to lock the best AI coding agent behind a paywall. Someone rebuilt the entire Claude Code from scratch in pure Python... ...and released it on GitHub for free. → Works with GPT, Gemini, DeepSeek, GLM, and more → Scores 58.2% on SWE-bench Verified → Around 6× cheaper to run → 180,000+ lines of production-ready code 100% Open Source. 👇 Link in the next post.
- 19
@KanikaBK ·
I write Python scripts and automation tooling and use Claude Code as my main assistant. Sessions involve a lot of file reading, editing, and testing. token costs were higher than they should be. Installed WozCode this week after seeing it on GitHub. the efficiency improvement on Python scripting work is noticeable. Vanilla Claude Code reads files, edits them, reads related files as three separate tool calls. Each call adds to context. WozCode batches them into one or two calls. across a session building a complex automation script, that difference compounds significantly. ran /woz-savings on my history. the number confirmed what I was feeling in the sessions. Cost is down. Sessions stay faster. This is a permanent addition to my setup.
Watch video
- 20
- 21
@milan_milanovic ·
𝗪𝗵𝗮𝘁 𝗵𝗮𝗽𝗽𝗲𝗻𝘀 𝘄𝗵𝗲𝗻 𝘆𝗼𝘂 𝗹𝗲𝘁 𝗖𝗹𝗮𝘂𝗱𝗲 𝗖𝗼𝗱𝗲 𝗽𝗶𝗰𝗸 𝘆𝗼𝘂𝗿 𝘁𝗼𝗼𝗹𝘀 𝗳𝗼𝗿 𝘆𝗼𝘂? Researchers sent 2,430 open-ended prompts to Claude Code across 3 models, 4 project types, and 20 categories. They did not mention any tools; they just asked, "What should I use?" Here is what they found: 𝟭. 𝗕𝘂𝗶𝗹𝗱 𝗼𝘃𝗲𝗿 𝗯𝘂𝘆 𝗶𝘀 𝘁𝗵𝗲 𝗱𝗲𝗳𝗮𝘂𝗹𝘁 Custom/DIY is the single most common "recommendation" in the dataset, 252 picks across 12 of 20 categories. Ask Claude Code to add feature flags, and it builds a system with env vars and React Context. Ask it to add auth to a Python project, and it writes a JWT from scratch every single time. When an agent can build a working solution in 30 seconds, it often does. 𝟮. 𝗔 𝗱𝗲𝗳𝗮𝘂𝗹𝘁 𝘀𝘁𝗮𝗰𝗸 𝗲𝘅𝗶𝘀𝘁𝘀 Where Claude Code does pick third-party tools, it converges hard: - GitHub Actions owns CI/CD at 94% - Stripe owns payments at 91% - shadcn/ui owns UI components at 90% - Vercel is a must for JavaScript projects at 100%. The rest of the list: PostgreSQL, Tailwind CSS, Zustand, pnpm, Resend, Vitest. These tools may not be the best option for your project, but these are what the model will choose for you. 𝟯. 𝗥𝗲𝗱𝘂𝘅 𝗶𝘀 𝗱𝗲𝗮𝗱 𝗶𝗻 𝗔𝗜-𝗮𝘀𝘀𝗶𝘀𝘁𝗲𝗱 𝗰𝗼𝗱𝗲 Redux did't got any primary picks across 2,430 prompts. The model knows it exists, with 23 mentions and 2 alternative recommendations, but never actually chooses it. Zustand wins state management at 65% instead. Express has it even worse. It doesn't show up as a primary pick, an alternative, or even a passing suggestion. It's just gone. 𝟰. 𝗡𝗲𝘄𝗲𝗿 𝗺𝗼𝗱𝗲𝗹𝘀 𝗽𝗿𝗲𝗳𝗲𝗿 𝗻𝗲𝘄𝗲𝗿 𝘁𝗼𝗼𝗹𝘀 This is the clearest signal from this dataset. Prisma goes from 79% in Sonnet 4.5 to 0% in Opus 4.6. Drizzle takes over completely. In Python projects, Celery usage collapses from 100% to 0% as newer models prefer FastAPI's built-in background tasks. It tracks with what appeared in more recent training data. 𝟱. 𝗖𝗼𝗻𝘁𝗲𝘅𝘁-𝗮𝘄𝗮𝗿𝗲𝗻𝗲𝘀𝘀 𝗶𝘀 𝗿𝗲𝗮𝗹 The same model picks Vercel for JavaScript and Railway for Python. Drizzle for Next.js, SQLModel for FastAPI. It's not a fixed list. The agent reads the stack and adapts, which is more useful than a blanket recommendation. 𝟲. 𝗕𝗲𝗶𝗻𝗴 𝗮𝗯𝘀𝗲𝗻𝘁 𝗳𝗿𝗼𝗺 𝗽𝗿𝗶𝗺𝗮𝗿𝘆 𝗽𝗶𝗰𝗸𝘀 𝗶𝘀𝗻'𝘁 𝘁𝗵𝗲 𝘀𝗮𝗺𝗲 𝗮𝘀 𝗯𝗲𝗶𝗻𝗴 𝗶𝗻𝘃𝗶𝘀𝗶𝗯𝗹𝗲 Netlify, SendGrid, and Jest were never chosen as the primary option. But they kept showing up as second choices. The model knows these tools and still recommends something else first. That gap is the one worth closing. If we're using AI coding agents for greenfield projects, we're increasingly inheriting a default stack. Worth knowing what that stack is. Full report in comments
- 22
@burkov ·
This ICLR 2025 paper documents OpenHands, a software platform that lets an AI agent operate a computer the way a developer does: writing and running code, issuing shell commands, and navigating web pages inside an isolated container. The technical core is an event stream, which is simply a running log of every action the agent takes and every observation it gets back; the agent reads this history at each step and decides what to do next, so building a new agent reduces to writing one function that maps the current history to the next action. Rather than giving the agent a fixed menu of tools, the design lets it express any action as ordinary Python or bash, which means new capabilities can be added as plain Python functions instead of being baked into the framework. The authors also wire in fifteen existing evaluation benchmarks covering bug fixing, real GitHub issue resolution, web navigation, and tool use, and report how the same generalist agent does across all of them without per-task tuning. The paper provides a concrete, implementation-level picture of how a code-executing agent is actually put together, including the parts usually left vague: how execution is sandboxed, how one agent hands a subtask to another, and how the team keeps agent behavior from silently regressing by recording and replaying model responses as deterministic tests. Read with an AI tutor: https://t.co/9fyKXu9fQL PDF: https://t.co/CnnX565DbX
- 23
@KirkDBorne ·
Competitive Programming in Python - 128 Algorithms to Develop Your Coding Skills: https://t.co/j4cELcvbjG "Classic problems like Dijkstra's shortest path algorithm and Knuth-Morris-Pratt's string matching algorithm are featured alongside lesser known data structures like Fenwick trees and Knuth's dancing links. The book provides a framework to tackle algorithmic problem solving, including: Definition, Complexity, Applications, Algorithm, Key Information, Implementation, Variants, In Practice, and Problems. Python code included in the book and on the companion website." -Amazon summary
- 24
@VaibhavSisinty ·
OpenAI didn’t just acquire a startup. They acquired the muscle memory of 10 million Python developers. Astral, the team behind uv, Ruff, and ty is now inside Codex. The default toolchain of Python developers. Let that sink in. This isn’t another “AI writes code” update. This is AI owning the entire development loop. → Plan the change → Spin up environments with uv → Type-check with ty → Lint with Ruff → Ship All… without breaking flow. And the smartest part? They’re keeping everything open-source
- 25
@tetsuoai ·
https://t.co/66cxhv7svg The X API is now available worldwide with pay-per-use pricing. X has also released official Python and TypeScript XDKs, an MCP server, xurl for agents, and a local Playground that lets you test the X API without using credits. If you spend on X API credits, you can get up to 20% back in XAI API credits. X is uniting real-time data, execution, testing, and developer tools into a single platform. This means X is becoming the go-to platform for agents.
- 26
@omarsar0 ·
Super interesting new work from NVIDIA. (bookmark it) They suggest building agents as Python objects. Very cool idea and I think it could a lot with agent reliability. More below: Agent development today spreads across prompt templates, tool schemas, callback code, and workflow graphs. NOOA replaces all four with one abstraction. An agent is a Python object. Its methods are the actions the model can take, its fields hold state, its docstrings are the prompts, and its type annotations act as contracts. A method whose body is "..." gets completed at runtime by a validated LLM loop. A method with a normal body stays deterministic Python. That single convention puts the boundary between probabilistic and deterministic behavior right in the source. Agent behavior becomes testable, traceable, and refactorable with the same tools you already use on the rest of your codebase. NVIDIA reports six model-facing ideas combined on one surface, including pass-by-reference over live objects and model-callable harness APIs for context and events, evaluated on SWE-bench Verified, Terminal-Bench 2.0, and ARC-AGI-3. Paper: https://t.co/PCtFtVY8rT Learn to build effective AI agents in our academy: https://t.co/1e8RZKs4uX
- 27
@agenticasdk ·
Many people have asked us: what changes when an agent has access to a persistent Python runtime? We ran a side-by-side comparison to demonstrate: Agentica's Python REPL-based agent vs traditional tool calling agents Full breakdown below 👇
Watch video
- 28
@_vmlops ·
OPENAI'S PYTHON LIBRARY HAS FEATURES MOST DEVS NEVER USE Most people just call the api and move on...but there's a lot more under the hood It handles retries, streaming, pagination & async out of the box so you're not writing boilerplate for things that should already work workload identity auth for k8s, azure, and gcp means no hardcoded api keys in production...webhook verification is built right in pip install openai https://t.co/k513J8OZTE
- 29
@MattNiessner ·
Some nostalgia: back in 2010 during my PhD, I interned at Microsoft Research, arguably the premier industry lab for academic research at the time. The role forced a switch from Debian to Windows, but the clear payoff was Visual Studio. Despite the protests of the Vim and Emacs purists, nothing on Linux came close to VS as an IDE. At the same time, IntelliSense, VS's early 'coding assistant' barely functioned. It routinely failed to parse multi-class C++ codebases, and the moment it encountered a complex expression template, it completely just gave up. A PhD colleague then recommended a simple plugin called Visual Assist. For me it was a revelation: a genuinely working implementation of context-aware completion and editing. It actually understood C++ syntax and its complexities, facilitating large-scale refactoring and reliable auto-complete. Then came the deep learning boom, and Python took over. CPU-side run-time as secondary when heavy parallel workloads were efficiently wrapped for the GPU. The abstraction advantage over C++ made Python the undisputed language for ML, and productivity soared as IDE integration and package management matured. In 2023, the next paradigm shift was LLM-based coding. Copilot institutionalized the 'tab-tab-tab' workflow with code prediction vastly superior to any prior heuristic. But even that feels primitive compared to the recent wave of coding agents. Just a year ago, maintaining reasonable productivity with these agents was a struggle; today, models like Claude Opus 4.6 operate at an entirely different level. It is a complete game-changer. Even senior researchers as myself can materialize complex projects... within hours :) There is zero doubt that agentic coding will soon become the absolute core of every SWE workflow. For me, an open question is what the correct abstraction and interface will look like. Generating high-level plans is one thing, but fully automating the implementation process while retaining strict user alignment remains a challenge.
- 30
- 31
@DivyanshT91162 ·
🤯 SQLModel eliminates one of the most annoying parts of Python development. No more maintaining separate Pydantic schemas and SQLAlchemy models. Define everything once with Python type annotations and get validation, serialization, and database functionality in a single model. Built on top of Pydantic and SQLAlchemy, with first-class FastAPI support and far less boilerplate than traditional setups. Clean, simple, and surprisingly powerful. Repo👇
- 32
@AskMichaelTaiwo ·
Satya Nadella said something more useful than most AI predictions, and it got buried. Asked about AI writing Microsoft's code, he noted the quality depends heavily on the language: the AI produces "fantastic" Python and is "not that great" at C++. One sentence, and it quietly explains the entire gap between AI demos and AI reality. Why would a machine be brilliant at one language and mediocre at another? Because it learned from what humans have publicly written. Python is everywhere online, tutorials, forums, millions of beginner projects. C++ is older, harder, and much of the best of it sits locked inside proprietary systems that never touched the open internet. AI is not smart in some general, even way. It is smart exactly where humanity left it a large, clean trail of examples, and dim where we didn't. This is the practical key to using these tools well, and almost nobody states it plainly. AI is strongest on the common, the documented, the done-a-million-times. It is weakest on the rare, the proprietary, the genuinely novel. So it will happily write your standard login page and struggle with the strange, specific edge case that is actually your business. Which tells you exactly where the human value moved. Not to the tasks AI does fantastically, those are commoditising toward zero as you read this. The value moved to the tasks AI is "not that great" at: the uncommon problem, the messy real-world context, the thing with no tidy corpus to learn from. That is the C++ of your own field, whatever your field is. The winners in this shift won't be the people who can do what AI already does well. They'll be the ones who figured out where the training data runs out, and made sure they live there.
- 33
- 34
@pycharm ·
How do you actually learn #Python? Mark Smith (@judy2k) breaks it down into 3 core ideas: 1. Start by copying At the beginning, you need guidance. Tutorials, books, and exercises help you understand how code is structured. 2. Move quickly to building This is where real learning happens. • Start small • Recreate simple tools • Work on projects you care about You don’t learn programming by consuming content – you learn it by writing code. 3. Learn from others Read code, follow experienced developers, and learn how real projects are written. ❌ Avoid common mistakes: - When you let AI write code, you stop learning. - When you try to learn everything at once, you risk burning out. - When you obsess about avoiding errors, you lose a major part of the process. ✅ What actually helps: - Breaking problems into small pieces. - Building early, even if it’s messy. - Learning gradually (including tools like Git). - Using AI to assist, not replace thinking. The key idea: You don’t learn programming by consuming content. You learn it by writing code. 👉 Watch the full talk: https://t.co/WFAmWvxtsZ
Watch video
- 35
- 36
@agenticgirl ·
Top 9 Python Libraries for 2026 Supercharge your workflow & stay ahead! 1. Polars → Rust-powered DataFrame → Insanely fast vs Pandas https://t.co/sHT1LsYxDG 2. Ruff → Lints + formats everything → Replaces 3 tools in one https://t.co/siO1xaOHO8 3. PyScript → Run Python in the browser → No backend needed https://t.co/YlrjYIn8mI 4. Pandera → Schema validation for data → Catch bugs early https://t.co/wZQ8aymtXg 5. JAX → Auto-diff + GPU/TPU → Built for high-performance ML https://t.co/XjbuUMq9b7 6. Textual → Build beautiful terminal apps → No frontend required https://t.co/f5Ax5ayUEb 7. LlamaIndex → Backbone for RAG apps → Connect LLMs to your data https://t.co/CMvqECj1Lu 8. Robyn → Async + Rust optimized → Blazing fast APIs https://t.co/dP3dEvd55o 9. DuckDB → In-memory analytics DB → Faster than SQLite https://t.co/TpKqrYaNPc
- 37
- 38
@s_gruppetta ·
There's never been a better time to learn core Python. Sure, AI is writing code for us now. But you still need to understand it, review it, guide the AI to what you really want English is great but can be ambiguous. A glance at the Python code helps you ensure the AI understood your requirements, or quickly steer it in the right direction. Python fundamentals take you further today than they ever did in the past. Beware of skipping the basics…
- 39
@MaheshPawaar ·
🐍 5 python concepts that made everything click for me: (bookmark this🔖) > generators – stops loading everything into memory at once – processes one item at a time with yield – saved me in django when querying large datasets > decorators – a function that wraps another function – that's literally all [@]login_required is – write your own for logging, auth, timing > context managers – "with" guarantees cleanup even when things break – calls `__enter__` and `__exit__` under the hood – django's transaction.atomic() is this exact pattern > list comprehensions – not just shorter syntax – shifts your thinking from "mutate step by step" to "describe what you want" – flatter code, easier to debug > unpacking – swap variables without a temp variable – destructure API responses in one line – grab first and last from a list cleanly – wish someone had told me sooner frameworks are just these stitched together. go deeper on what you have. not wider.
- 40
- 41
- 42
- 43
- 44
@__mharrison__ ·
I had the chance this week to teach a Professional Python class to a room full of very smart people. My IQ was definitely bringing the average down. Almost everyone in the room had a PhD. Except for two of us. One of them was me. And I saw something I often see when I work with highly technical experts. They are brilliant. They know their domains deeply. They know how to get things done. They have built workflows that work for them, often over many years. But there are usually gaps around coding and software engineering practices. Their job has usually been to get the analysis working, get the model running, get the paper out, get the result shipped, or get the thing working on their machine. That creates a very different coding style than what you need when code needs to be shared, tested, refactored, reviewed, maintained, and collaborated on. That is the fun part of teaching this material. You start introducing processes and practices that move people from being highly capable individuals to being a team that can build together. Testing was one of the big unlocks. Many people had not spent much time with it, but they immediately saw its value. Tests give you confidence. They make refactoring safer. They make collaboration easier. They let you change code without holding your breath. For the skeptics, I told them to apply the scientific process (they are all scientists, so they can't do much to argue), apply the techniques, and verify whether the results are better. They were also very interested in AI-assisted coding. This class was not an AI coding class. I teach that separately. But this class is foundational for AI coding because AI coding accelerates experts. The better your software development practices are, the better your results with AI will be. If you bring messy habits to AI, you get faster messes. If you bring testing, structure, environments, refactoring, and collaboration practices, AI becomes much more useful. Professional software practices are not just for software engineers anymore. They are the foundation for anyone who wants to use Python, data, and AI at a high level.
- 45
@ATechAjay ·
Python Roadmap for Frontend Engineers. Step 1: Python Fundamentals □ Syntax & Basics □ Data Types, Variables & Operators □ Control Flow & Loops □ Functions & Modules □ OOP (Classes & Inheritance) □ Error Handling & File I/O Step 2: Data Handling & Visualization □ NumPy & Pandas □ Matplotlib & Seaborn □ Plotly (Interactive Charts) □ Exploratory Data Analysis (EDA) Step 3: Traditional Full-Stack Backend □ Virtual Environments □ Flask or FastAPI □ Django □ REST APIs & Authentication □ Connect to your existing JS/React Frontends Step 4: Pure Python UI Frameworks □ Streamlit – Fast Data Apps & Prototypes □ Gradio – ML Model Interfaces & Demos □ NiceGUI – Modern UI with Great Styling □ Reflex – Full-Stack Apps (React under the hood, all Python) □ Dash / Panel – Advanced Interactive Dashboards Step 5: Styling & Advanced UI □ Custom CSS, Themes & Layouts □ Component Libraries & Extensions □ State Management & Reactivity □ Responsive Design Patterns □ Blend with your HTML/CSS/JS Expertise Step 6: Databases & State Management □ SQLite / PostgreSQL □ ORM Basics (SQLAlchemy) □ Real-time Updates & WebSockets □ Caching & Session Handling Step 7: AI-Powered Interfaces □ OpenAI & Hugging Face API Integration □ LangChain / LlamaIndex for Smart UIs □ Build Chatbots, RAG Apps & Agents □ Gradio + LLMs or Streamlit + AI Features Step 8: Deployment & MLOps □ Hugging Face Spaces □ Streamlit Cloud, Render, Vercel □ Docker Basics for Python Apps □ CI/CD & GitHub Actions Step 9: Version Control & Best Practices □ Git & GitHub □ Clean Project Structure □ Testing UI Components □ Performance Optimization Step 10: Build Projects □ Interactive Data Dashboard (Streamlit/Dash) □ AI Chatbot or RAG Interface (Gradio/Reflex) □ Full-Stack Web App (Reflex or FastAPI + React) □ ML Model Demo Platform □ Deployed Portfolio Site or SaaS Prototype Now start building Python-powered UIs and apps. Your frontend experience gives you a huge head start in design, UX & interactivity! Good luck 🚀
- 46
- 47
@jianw851 ·
Most people use 5+ paid tools or build a web app for this. But in #OpenClaw era it could be as simple as a python file: A full spaced-repetition study tracker: • pattern analysis • review scheduling • drill management Zero dependencies. Zero setup. No web app. Pure CLI. Want the Code? comment bellow. python tasks/cli.py add # log a task python tasks/cli.py todo # what's due today python tasks/cli.py review # active recall python tasks/cli.py stats # see dominant patterns python tasks/cli.py drills # targeted micro-practice SM-2 scheduling built in. Intervals grow as you improve, reset when you blank. One JSON file as the database. And because it's just a CLI over structured data — an AI agent can drive it directly. No API to wrap, no UI to scrape. add, review, stats are already agent-friendly commands. The whole thing fits in your repo.
- 48
- 49
- 50
Best Tweets by Topic