Eric Ma
@ericmjl.bsky.social
Senior Principal Data Scientist, Moderna. Trying to follow Jesus.
Coding agents can wreck your repo in seconds. Auto-approve only read-only commands—never risky ones. I share my setup for safe agent autonomy and lessons learned. How do you keep agents in check? Comment or repost! buff.ly/PetnnVz #ai #automation
Safe ways to let your coding agent work autonomously
written by Eric J. Ma on 2025-11-08 | tags: automation productivity coding agents safety workflow development prompting command line ai
buff.ly
November 12, 2025 at 3:03 PM
Coding agents can wreck your repo in seconds. Auto-approve only read-only commands—never risky ones. I share my setup for safe agent autonomy and lessons learned. How do you keep agents in check? Comment or repost! buff.ly/PetnnVz #ai #automation
AI agents + Marimo notebooks = Python on autopilot. Letting AI write and check my Marimo code is a total workflow upgrade. The --watch flag and marimo check catch errors fast and keep me moving. Would you try this? Repost or comment! buff.ly/6LDN1RV #python #datascience
November 5, 2025 at 3:05 PM
AI agents + Marimo notebooks = Python on autopilot. Letting AI write and check my Marimo code is a total workflow upgrade. The --watch flag and marimo check catch errors fast and keep me moving. Would you try this? Repost or comment! buff.ly/6LDN1RV #python #datascience
By definition, AI agents are always going to be out of date. AI agents need live, searchable knowledge. I share how LlamaBot's MCP server keeps docs fresh and accessible for any LLM agent. How can this help you? Repost or comment! buff.ly/5td8FNC #ai #documentation
How to expose any documentation to any LLM agent
written by Eric J. Ma on 2025-10-19 | tags: llm documentation ai mcp workflow context search knowledge development automation
buff.ly
October 29, 2025 at 2:05 PM
By definition, AI agents are always going to be out of date. AI agents need live, searchable knowledge. I share how LlamaBot's MCP server keeps docs fresh and accessible for any LLM agent. How can this help you? Repost or comment! buff.ly/5td8FNC #ai #documentation
Skills cut token costs—MCP servers win for sharing. Which is better for LLMs? Skills = fast, local tweaks. MCP = team standard. I compared Anthropic's skills repo to MCP servers—real trade-offs, not just hype. Which would you use? Repost or comment! buff.ly/zhV1XfO #llm #aiworkflows
Exploring Skills vs MCP Servers
written by Eric J. Ma on 2025-10-20 | tags: anthropic skills token efficiency llm automation customization workflows development mcp
buff.ly
October 22, 2025 at 2:06 PM
Skills cut token costs—MCP servers win for sharing. Which is better for LLMs? Skills = fast, local tweaks. MCP = team standard. I compared Anthropic's skills repo to MCP servers—real trade-offs, not just hype. Which would you use? Repost or comment! buff.ly/zhV1XfO #llm #aiworkflows
Coding agents flop without real workflows—prompts aren't enough. Discipline and feedback loops make agents useful. TDD-first, external memory, and fast iteration changed my game. What’s your best agent tip? Comment or share! buff.ly/yH5Hjkv #aicoding #devworkflow
October 15, 2025 at 2:01 PM
Coding agents flop without real workflows—prompts aren't enough. Discipline and feedback loops make agents useful. TDD-first, external memory, and fast iteration changed my game. What’s your best agent tip? Comment or share! buff.ly/yH5Hjkv #aicoding #devworkflow
What if your coding agent could remember your preferences? AGENTS.md lets you train your LLM agent like an employee. This one file made my workflow smoother—no more repeating myself! Would you try this? Repost or comment! buff.ly/LinjaUs #.aiagents #.devtips
October 8, 2025 at 2:06 PM
What if your coding agent could remember your preferences? AGENTS.md lets you train your LLM agent like an employee. This one file made my workflow smoother—no more repeating myself! Would you try this? Repost or comment! buff.ly/LinjaUs #.aiagents #.devtips
Ultralearning is a superpower to mastering tough skills quickly. Courses aren't enough—self-driven ultralearning is the real unlock. I break down how ultralearning helps you grow fast, whether it's bio, software, or any field. What will you ultralearn next? buff.ly/2l4z68Z #ultralearning #growth
How data scientists can master life sciences and software skills for biotech using ultralearning
written by Eric J. Ma on 2025-10-01 | tags: biotech ultralearning datascience lifesciences software learning career skills modeling feedback
buff.ly
October 2, 2025 at 2:02 PM
Ultralearning is a superpower to mastering tough skills quickly. Courses aren't enough—self-driven ultralearning is the real unlock. I break down how ultralearning helps you grow fast, whether it's bio, software, or any field. What will you ultralearn next? buff.ly/2l4z68Z #ultralearning #growth
Still fighting Python environment headaches? I’ve moved to pixi—automation and new tools are the future. My Data Science Bootstrap Notes now use pixi, uv, and AI. This is my recommended workflow for 2025. Would you make the switch? Comment or repost! buff.ly/MoXJJhL #.datascience #.python
September 24, 2025 at 12:20 PM
Still fighting Python environment headaches? I’ve moved to pixi—automation and new tools are the future. My Data Science Bootstrap Notes now use pixi, uv, and AI. This is my recommended workflow for 2025. Would you make the switch? Comment or repost! buff.ly/MoXJJhL #.datascience #.python
AI isn’t just for code—here’s how I use it to boost my career. LLMs help me prep for negotiations, updates, and tough convos. 10 ways I save time and stand out at work, with prompts you can steal. Which would you try? Comment or repost! buff.ly/OkXup6p #aihacks #careerdevelopment
How to use AI to accelerate your career in 2025
written by Eric J. Ma on 2025-09-01 | tags: productivity negotiation presentations llm automation communication competencies ghostwriting updates ai
buff.ly
September 17, 2025 at 11:02 AM
AI isn’t just for code—here’s how I use it to boost my career. LLMs help me prep for negotiations, updates, and tough convos. 10 ways I save time and stand out at work, with prompts you can steal. Which would you try? Comment or repost! buff.ly/OkXup6p #aihacks #careerdevelopment
Ever wonder why your stats brilliance doesn’t land with lab scientists? Clarity beats complexity every time. Here’s how I learned to lead with decisions, not methods. What’s your best tip for bridging the gap? buff.ly/8Zuli8E #biotech #datatips
How to communicate with lab scientists (when you're the data person)
written by Eric J. Ma on 2025-08-24 | tags: biotech communication decisions statistics translation collaboration trust meetings probability stakeholders
buff.ly
September 10, 2025 at 11:02 AM
Ever wonder why your stats brilliance doesn’t land with lab scientists? Clarity beats complexity every time. Here’s how I learned to lead with decisions, not methods. What’s your best tip for bridging the gap? buff.ly/8Zuli8E #biotech #datatips
🔥 You can rewrite Python functions live! It’s wild, risky, and opens up new AI tricks. I tried dynamic patching—super powerful, but security is a real concern. Here’s what I learned. Would you try this, or is it too risky? buff.ly/RqGBgE1 #python #ai
Wicked Python trickery - dynamically patch a Python function's source code at runtime
written by Eric J. Ma on 2025-08-23 | tags: python runtime llm security namespace compilation execution functions toolbot monkeypatching
buff.ly
September 3, 2025 at 11:02 AM
🔥 You can rewrite Python functions live! It’s wild, risky, and opens up new AI tricks. I tried dynamic patching—super powerful, but security is a real concern. Here’s what I learned. Would you try this, or is it too risky? buff.ly/RqGBgE1 #python #ai
Are data scientists obsolete now that LLMs can code? I see LLMs as tools that reshape, not replace, our work. Experimenting with LLMs has made our scientific mindset more valuable than ever. What’s your take? Repost or comment! buff.ly/o9PnXhR #llms #datascience
Data scientists aren't becoming obsolete in the LLM era
written by Eric J. Ma on 2025-08-15 | tags: productivity workflows evaluation metrics business science models ai tools measurement
buff.ly
August 27, 2025 at 11:02 AM
Are data scientists obsolete now that LLMs can code? I see LLMs as tools that reshape, not replace, our work. Experimenting with LLMs has made our scientific mindset more valuable than ever. What’s your take? Repost or comment! buff.ly/o9PnXhR #llms #datascience
Reposted by Eric Ma
I've been eval pilled by @hamel.bsky.social . Everyone is all let's build some MCP servers and ship a minimal AISRE as quickly as possible. And I'm writing a design doc about how to build evals with @runme.dev so we can iterate rapidly on the AI
August 8, 2025 at 2:54 PM
I've been eval pilled by @hamel.bsky.social . Everyone is all let's build some MCP servers and ship a minimal AISRE as quickly as possible. And I'm writing a design doc about how to build evals with @runme.dev so we can iterate rapidly on the AI
Reposted by Eric Ma
RFdiffusion2 is now live!
github.com/RosettaCommo...
You can now design proteins, and in particular enzymes from just partially defined amino acid side chains, and without defining their sequence position or order!
github.com/RosettaCommo...
You can now design proteins, and in particular enzymes from just partially defined amino acid side chains, and without defining their sequence position or order!
August 22, 2025 at 12:51 AM
RFdiffusion2 is now live!
github.com/RosettaCommo...
You can now design proteins, and in particular enzymes from just partially defined amino acid side chains, and without defining their sequence position or order!
github.com/RosettaCommo...
You can now design proteins, and in particular enzymes from just partially defined amino acid side chains, and without defining their sequence position or order!
Stop guessing priors—R2D2 puts you in control. I think R2D2 is a breakthrough for Bayes modeling. It automates shrinkage and variance allocation for clearer, less overfit models. Would you try this? Repost or comment! buff.ly/kcbOuG6 #bayesianstats #statmodeling
Stop guessing at priors: R2D2's automated approach to Bayesian modeling
written by Eric J. Ma on 2025-08-06 | tags: bayesian variance r2d2 dirichlet multilevel glm regularization priors inference pymc
buff.ly
August 20, 2025 at 11:02 AM
Stop guessing priors—R2D2 puts you in control. I think R2D2 is a breakthrough for Bayes modeling. It automates shrinkage and variance allocation for clearer, less overfit models. Would you try this? Repost or comment! buff.ly/kcbOuG6 #bayesianstats #statmodeling
AI can't replace the grind—you have to earn your automation. My 2-day feature rewrite only worked because of months of prep. I share how struggle made AI a true thinking partner, not a shortcut. Have you tried this approach? Repost or comment! buff.ly/TG4xOqr #.ai #.softwaredev
From nerd-sniped to shipped using AI as a thinking tool
written by Eric J. Ma on 2025-07-21 | tags: automation ai memory design coding testing architecture prototyping review pairing
buff.ly
August 13, 2025 at 11:02 AM
AI can't replace the grind—you have to earn your automation. My 2-day feature rewrite only worked because of months of prep. I share how struggle made AI a true thinking partner, not a shortcut. Have you tried this approach? Repost or comment! buff.ly/TG4xOqr #.ai #.softwaredev
Reposted by Eric Ma
🧵Americans should really hear how the CDC in other regions talks about us.
August 9, 2025 at 3:06 AM
🧵Americans should really hear how the CDC in other regions talks about us.
Reposted by Eric Ma
They spread the lie for years that public health professionals are enemies. Now, a policeman is dead, and the lives of CDC personnel and their kids were at risk.
There'll be "deep concern" and thoughts & prayers.
But architects of that lie will celebrate privately because this has been the goal.
There'll be "deep concern" and thoughts & prayers.
But architects of that lie will celebrate privately because this has been the goal.
Police officer dead after ‘antivax’ shooter targets CDC headquarters
An active shooter was reported on the Emory University campus in Georgia. The suspect, who was killed, was said to have blamed his illness on the Covid vaccine
www.thetimes.com
August 9, 2025 at 3:26 AM
They spread the lie for years that public health professionals are enemies. Now, a policeman is dead, and the lives of CDC personnel and their kids were at risk.
There'll be "deep concern" and thoughts & prayers.
But architects of that lie will celebrate privately because this has been the goal.
There'll be "deep concern" and thoughts & prayers.
But architects of that lie will celebrate privately because this has been the goal.
Reposted by Eric Ma
CDC Buildings Hit by Gunfire :
MAGA’s war on truth works exactly as designed...turn Americans against science, poison them with propaganda, and watch them die from preventable causes… all while calling it “freedom.” This is the bill for their lies.
www.medpagetoday.com/publichealth...
MAGA’s war on truth works exactly as designed...turn Americans against science, poison them with propaganda, and watch them die from preventable causes… all while calling it “freedom.” This is the bill for their lies.
www.medpagetoday.com/publichealth...
CDC Buildings Hit by Gunfire
Agency may have been the target of shooter who blamed his illness on COVID shots
www.medpagetoday.com
August 9, 2025 at 3:19 AM
CDC Buildings Hit by Gunfire :
MAGA’s war on truth works exactly as designed...turn Americans against science, poison them with propaganda, and watch them die from preventable causes… all while calling it “freedom.” This is the bill for their lies.
www.medpagetoday.com/publichealth...
MAGA’s war on truth works exactly as designed...turn Americans against science, poison them with propaganda, and watch them die from preventable causes… all while calling it “freedom.” This is the bill for their lies.
www.medpagetoday.com/publichealth...
What if all your lab and ML data lived in one place? Index chaos is optional—xarray keeps everything aligned. I show how this cuts errors and makes analysis smoother. Would this help your workflow? Repost or comment! buff.ly/6QXFDY4 #xarray #datamanagement
How to use xarray for unified laboratory data storage
written by Eric J. Ma on 2025-07-15 | tags: xarray bioinformatics reproducibility cloud workflow alignment features laboratory datasets scaling
buff.ly
August 6, 2025 at 11:02 AM
What if all your lab and ML data lived in one place? Index chaos is optional—xarray keeps everything aligned. I show how this cuts errors and makes analysis smoother. Would this help your workflow? Repost or comment! buff.ly/6QXFDY4 #xarray #datamanagement
What keeps me coming back to #scipy after 10 years? The community’s curiosity and generosity are unmatched. From teaching to spontaneous podcast chats, every year brings new inspiration and growth. Got a favorite conference story? Comment or repost! buff.ly/gRAMkoC #scipy #python
Reflections on the SciPy 2025 Conference
written by Eric J. Ma on 2025-07-14 | tags: scipy python conference marimo tutorials llms xarray community networking career
buff.ly
July 30, 2025 at 11:02 AM
What keeps me coming back to #scipy after 10 years? The community’s curiosity and generosity are unmatched. From teaching to spontaneous podcast chats, every year brings new inspiration and growth. Got a favorite conference story? Comment or repost! buff.ly/gRAMkoC #scipy #python
Reposted by Eric Ma
Straight talk from Dr. Francis Collins on the devastating - and politically motivated - cuts to science and public health in the United States.
Anybody who cares about their health or the future or science, technology, and innovation in this country should watch this.
www.cnn.com/2025/07/25/T...
Anybody who cares about their health or the future or science, technology, and innovation in this country should watch this.
www.cnn.com/2025/07/25/T...
‘It’s heartless, it’s careless, and it’s deeply damaging’ says former NIH director on cuts to medical research | CNN
Dr. Francis Collins, former director of the National Institutes of Health, speaks to Christiane Amanpour about his decision to leave the agency and what he says is the Trump administration’s crackdown...
www.cnn.com
July 25, 2025 at 8:48 PM
Straight talk from Dr. Francis Collins on the devastating - and politically motivated - cuts to science and public health in the United States.
Anybody who cares about their health or the future or science, technology, and innovation in this country should watch this.
www.cnn.com/2025/07/25/T...
Anybody who cares about their health or the future or science, technology, and innovation in this country should watch this.
www.cnn.com/2025/07/25/T...
Can AI actually make us worse learners if we use it too soon? We should earn the privilege to use automation—basics first. A real story of AI in education gone wrong, and why process matters. Seen AI help or hurt learning? Share your thoughts! ericmjl.github.io/blog/2025/7/... #aiineducation
Earn the privilege to use automation
written by Eric J. Ma on 2025-07-13 | tags: education assessment automation ai learning workplace skills process outcomes privilege
ericmjl.github.io
July 23, 2025 at 11:02 AM
Can AI actually make us worse learners if we use it too soon? We should earn the privilege to use automation—basics first. A real story of AI in education gone wrong, and why process matters. Seen AI help or hurt learning? Share your thoughts! ericmjl.github.io/blog/2025/7/... #aiineducation
The real competition for your docs? It's not that strong. Even basic docs beat digging through Slack or bugging teammates. You don't need perfect docs—just make them the fastest way to get things done. Are your docs the easiest option? ericmjl.github.io/blog/2025/7/... #documentation #productivity
The job your docs need to do
written by Eric J. Ma on 2025-07-07 | tags: documentation diataxis innovation tutorials guides reference ai strategy product jobs theory
ericmjl.github.io
July 16, 2025 at 11:03 AM
The real competition for your docs? It's not that strong. Even basic docs beat digging through Slack or bugging teammates. You don't need perfect docs—just make them the fastest way to get things done. Are your docs the easiest option? ericmjl.github.io/blog/2025/7/... #documentation #productivity