Peter Plucinski

@mdancho84 Docling is great, especially running on GPU

🚨 BREAKING: IBM launches a free Python library that converts ANY document to data Introducing Docling. Here's what you need to know: 🧵

@func25 @MattJamesBoyle Go simplicity is its strong point. Almost on par with Rust in speed but way easier to read and write

Go is simple, so I ended up writing an 865-page book about how it works internally, just to see how it maintains that simplicity 😇

@_vmlops Its a great library but i prefer docling slightly over this, ideally docling running on gpu

MICROSOFT BUILT A TOOL THAT CONVERTS LITERALLY ANYTHING INTO CLEAN MARKDOWN FOR YOUR LLM pdfs. word docs. excel. powerpoint. audio. youtube urls one pip install and your AI pipeline stops choking on raw files forever no custom parsers. no broken layouts. no garbled text. just clean, structured markdown your LLM can actually read github.com/microsoft/mark…

@DougAillm @_vmlops Yes. Markitdown is a great library but have you tried docling as an alternative?

@_vmlops Been running this in agent pipelines for months. Great for structured docs - financials, spec sheets, Word. Weak spots: scanned PDFs (no OCR) and complex merged tables. For those, add a vision model pass. Best open-source option for the 80% case.

I still remember when people thought "prompt engineering" was going to become a real career.

@systemdesignone Head First Design Patterns

What's a software engineering book you wish you could read again for the very first time?

Most of the people who think that AI will replace developers are: - Managers who don’t code - Investors and startup founders selling it - People outside tech Developers: "It's helpful."

Linus Torvalds created Linux at 21 without Claude or any other AI. - He didn't have a co-founder. - No VC funding. No office. - No team. - Just a personal project he posted to a mailing list: "I'm doing a free OS." 33 years later, it runs 97% of the world's servers, all smartphones, and the International Space Station. The most important software in history started as someone's side project. Absolute legend.

Seems AWS has volunteered to be a crash test dummy for vibe coding. theguardian.com/technology/202…

Pure vibe coding is a Ponzi scheme. Eventually, the technical debt comes due, and if you don't understand the foundations, you can't pay it off.

People giving OpenClaw root access to their entire life

The real story is about what happens when you mandate agentic AI adoption before your guardrails exist. Amazon set an internal target of 80% of developers using AI coding tools weekly and tracked adoption closely. Leadership signed a memo pushing Kiro as the default for all production work. Engineers who wanted Claude Code instead needed VP-level approval. 1,500 employees petitioned against the policy. The company ignored them. Then Kiro got operator-level permissions with no mandatory peer review. An engineer let it resolve a production issue autonomously. The AI decided the best fix was to delete and recreate the entire environment. 13 hours of downtime on a system inside the division that generates 60% of Amazon’s operating profit. This was the second AI-caused production outage in months. Amazon Q Developer caused another one. Both times, the AI tools had the same permissions as human engineers but none of the institutional muscle memory that tells a senior dev “maybe don’t nuke the environment at 2pm on a Tuesday.” Amazon’s response tells you everything: “user error, not AI error.” They only added mandatory peer review and safety training after both incidents. The safeguards everyone assumed existed didn’t. And Amazon isn’t alone. Google’s Antigravity wiped a developer’s entire hard drive in December trying to clear a cache. Replit’s AI deleted a production database earlier in 2025 and then fabricated fake data to cover it up. Three different companies. Three different AI coding tools. Same failure pattern: agentic permissions without agentic guardrails. Google’s own 2025 DORA report found 90% of developers use AI for coding but only 24% trust it “a lot.” The adoption is running way ahead of the trust, and the trust is running way ahead of the infrastructure. The pattern across every one of these incidents is identical: company mandates AI adoption → sets aggressive usage targets → gives the tool production access → skips the review processes they’d require for any human engineer → acts surprised when the autonomous agent does something autonomously destructive. The question everyone keeps asking is whether AI can write code. The real question is whether organizations will build the permission structures, blast radius containment, and approval workflows before or after the outages force them to. Right now the answer is after. Every time.

Boris created Claude Code. His point here is important - when AI handles the code generation, the engineer's value shifts to the decisions above the code: 1. what do we build? 2. why? for whom? 3. and how it all fits together. The bottleneck was always judgment, taste, and systems thinking. AI just made that more obvious.

Yep... if you don’t understand how to write code yourself, you can’t evaluate what the AI gives you.

Vibe coding and trunk based development is going to destroy companies. 🍿

Most of the people who think that AI will replace developers are: - Managers who don’t code - Investors and startup founders selling it - OpenAI, Antropic, Google folks Developers: "It's helpful".

Big AI makes money selling models. They need you to believe that there's no hope unless you start using them. When you stop thinking for yourself, let your skills erode, and become entirely dependent on those models, they profit. They will say whatever supports this agenda.

I'll be impressed when you vibe code some paying customers.

100% I've always said the same. A startup with a couple of engineers should not be using complicated solutuons - Kubernates was designed to solve problems with massive scale, startups are nowhere near that scale

Bad news: AI coding tools don't work for business logic or with existing code, and can't replace domain knowledge or human decision-making. They're just good for boilerplate and simple, repetitive tasks. AGI is not at hand. arxiv.org/abs/2512.14012