Mark Finkle's Avatar

Mark Finkle

@mfinkle.bsky.social

Technology curmudgeon https://starkravingfinkle.org/blog/ https://mastodon.cloud/@mfinkle

158 Followers  |  74 Following  |  40 Posts  |  Joined: 21.11.2024  |  1.8929

Latest posts by mfinkle.bsky.social on Bluesky

Has anyone merged reactive notebooks like Marimo with conversational agents yet? I mean, can I ask some questions about data and the agent fetches the data and displays the results. I can then interactive with either the data or the results.

That might be neat

01.11.2025 03:51 โ€” ๐Ÿ‘ 0    ๐Ÿ” 0    ๐Ÿ’ฌ 1    ๐Ÿ“Œ 0

amusing too ๐Ÿซ 

01.11.2025 03:29 โ€” ๐Ÿ‘ 0    ๐Ÿ” 0    ๐Ÿ’ฌ 0    ๐Ÿ“Œ 0

"we're Chrome in a trenchcoat" โ€” this

"however it does give them an escape route if something can replace Chrome." โ€” I would not hold my breath. It's kinda assuming to realize that making a web browser is possibly harder than an LLM? ๐Ÿ˜‚

01.11.2025 03:27 โ€” ๐Ÿ‘ 2    ๐Ÿ” 0    ๐Ÿ’ฌ 1    ๐Ÿ“Œ 0
Post image

I didn't check to see if @siracusa.social was named in the article

17.10.2025 00:07 โ€” ๐Ÿ‘ 1    ๐Ÿ” 0    ๐Ÿ’ฌ 0    ๐Ÿ“Œ 0

Random weird thing: Apple Pay is only a direct payment option in Starbucks app without an account. When you create your account the option disappears... and the experience gets worse.

I deleted the account

13.10.2025 18:31 โ€” ๐Ÿ‘ 0    ๐Ÿ” 0    ๐Ÿ’ฌ 0    ๐Ÿ“Œ 0

I can't imagine AOL employees are happy about this outcome

02.10.2025 03:22 โ€” ๐Ÿ‘ 0    ๐Ÿ” 0    ๐Ÿ’ฌ 0    ๐Ÿ“Œ 0

Hard to imagine how the government being shut down could be much worse than the government weโ€™ve had over the past 9 months.

27.09.2025 13:52 โ€” ๐Ÿ‘ 802    ๐Ÿ” 112    ๐Ÿ’ฌ 36    ๐Ÿ“Œ 11
Post image 25.09.2025 21:43 โ€” ๐Ÿ‘ 1    ๐Ÿ” 0    ๐Ÿ’ฌ 0    ๐Ÿ“Œ 0
Video thumbnail

Dam been writing a book about this and she just TikToked it in 90 seconds.

19.09.2025 09:33 โ€” ๐Ÿ‘ 19670    ๐Ÿ” 8195    ๐Ÿ’ฌ 261    ๐Ÿ“Œ 765

๐Ÿ‘€ "the US experience is peculiar. Young Euro area workers have record low unemployment. In the UK, the young personsโ€™ unemployment rate has fallen steadily. Employment by young Japanese workers is near all-time highs. It seems implausible that AI uniquely hurts the prospects of younger US workers."

19.09.2025 11:28 โ€” ๐Ÿ‘ 59    ๐Ÿ” 17    ๐Ÿ’ฌ 2    ๐Ÿ“Œ 0

Amazing how those democratic lawmakers who were hunted down in their homes in the middle of the night in Minnesota werenโ€™t a watershed moment of political violence, huh?

11.09.2025 05:10 โ€” ๐Ÿ‘ 18666    ๐Ÿ” 5167    ๐Ÿ’ฌ 264    ๐Ÿ“Œ 101

enshittification | noun | when a digital platform is made worse for users, in order to increase profits

03.09.2025 20:22 โ€” ๐Ÿ‘ 29339    ๐Ÿ” 8644    ๐Ÿ’ฌ 512    ๐Ÿ“Œ 658

Yes! Although I didn't write it, I spent enough time in the code to appreciate it and notice when other products don't have it

21.08.2025 12:29 โ€” ๐Ÿ‘ 1    ๐Ÿ” 0    ๐Ÿ’ฌ 0    ๐Ÿ“Œ 0

I'm concerned that my first thought was:
Does @siracusa.social know about this?

14.08.2025 21:30 โ€” ๐Ÿ‘ 1    ๐Ÿ” 0    ๐Ÿ’ฌ 0    ๐Ÿ“Œ 0

Miss you too, but now we've found each other again!

27.07.2025 13:05 โ€” ๐Ÿ‘ 2    ๐Ÿ” 0    ๐Ÿ’ฌ 0    ๐Ÿ“Œ 0
Preview
Celebrating 20 years of MDN | MDN Blog MDN turns 20! Let's look at how we started, how MDN became the most trusted resource for web developers, the impact it's had on the open web, and yes, there's cake, too.

For 20 years Mozilla has developed one of the most useful tools for web devs.

developer.mozilla.org/en-US/blog/m...

Congrats to the MDN team on an awesome milestone. I'm happy that we've played a small part in its history and I'm excited to see it continue to be at the core of every web dev's job.

25.07.2025 07:58 โ€” ๐Ÿ‘ 140    ๐Ÿ” 46    ๐Ÿ’ฌ 0    ๐Ÿ“Œ 1

Pre-sleep thought: My hotel room has a mini-fridge, empty, but running.

Opportunity for removing mini-fridges to save energy? Or at least make a smarter mini-fridge?

21.07.2025 03:45 โ€” ๐Ÿ‘ 0    ๐Ÿ” 0    ๐Ÿ’ฌ 1    ๐Ÿ“Œ 0

Seen in Slack: it is working due to a litany of absolute HACKS!

One person's hacks is another person's temporary fix

17.07.2025 02:02 โ€” ๐Ÿ‘ 0    ๐Ÿ” 0    ๐Ÿ’ฌ 0    ๐Ÿ“Œ 0

I wonder that if, from James Martin's point of view in 1982, that actually happened? But from our 2025 point of view, the programmers moved to work on other things?

14.07.2025 22:14 โ€” ๐Ÿ‘ 0    ๐Ÿ” 0    ๐Ÿ’ฌ 0    ๐Ÿ“Œ 0
Preview
Figma files for IPO on NYSE, plans to 'take big swings' with acquisitions Figma filed its IPO prospectus on Tuesday, as the design software vendor looks to raise money and awareness to take on Adobe.

Congratulations to Figma for filing for an IPO. Regulators blocking the Adobe deal must have felt bad in the moment but this is a better outcome.

Stay independent and get rich on your own terms.

02.07.2025 01:12 โ€” ๐Ÿ‘ 39    ๐Ÿ” 3    ๐Ÿ’ฌ 3    ๐Ÿ“Œ 2
Preview
E.P.A. Plans to Reconsider a Ban on Cancer-Causing Asbestos

"We're all going to die anyway" - Ancient Republican Proverb

18.06.2025 12:53 โ€” ๐Ÿ‘ 77    ๐Ÿ” 17    ๐Ÿ’ฌ 2    ๐Ÿ“Œ 0
Preview
Perplexity CEO says its browser will track everything users do online to sell 'hyper personalized' ads | TechCrunch Perplexity is building its own browser is to collect data on everything users do outside of its own app to sell ads.

What reason could I possibly have to use such a browser?
techcrunch.com/2025/04/24/p...

01.05.2025 18:31 โ€” ๐Ÿ‘ 0    ๐Ÿ” 0    ๐Ÿ’ฌ 1    ๐Ÿ“Œ 0
Exploring LLMs as Agents: Google Agent Dev Kit Up to this point, I have been experimenting with my own home-grown agent framework, based on Simon Willisonโ€™s great LLM project to handle wrapping the different LLMs. Armed with some experienโ€ฆ

Google released Agent Development Kit, a framework for building agent workflows with everything we expect from an agent frameworks.

I recreated my ToolAgent quasi-framework using ADK. I could expose my mock tools, add ReAct planning, and switch between different LLMs. It worked out really well. ๐ŸŽ‰

14.04.2025 20:34 โ€” ๐Ÿ‘ 0    ๐Ÿ” 0    ๐Ÿ’ฌ 0    ๐Ÿ“Œ 0
Exploring LLMs as Agents: Local Models I finally got around to exploring local models, which is surprisingly simple to set up. I wanted to see how well a local model would perform in ToolAgent compared to the remote models (mostly Geminโ€ฆ

I finally got around to exploring local LLMs. It was easier to set up than I expected. I tired a few different models of different sizes. The smaller ones didn't have enough "reasoning" ability to work as agents, but the larger ones did a pretty good job.
starkravingfinkle.org/blog/2025/04...

14.04.2025 13:45 โ€” ๐Ÿ‘ 0    ๐Ÿ” 0    ๐Ÿ’ฌ 0    ๐Ÿ“Œ 0
Preview
Exploring LLMs as Agents: Model Context Protocol I will likely run out of topics to cover in the โ€œLLMs as Agentsโ€ series, eventually. I wanted to explore creating a chat UI and what I could do with embeddings but I didnโ€™t do thaโ€ฆ

I took some time to experiment with MCP Client and Servers. I converted my ToolProvider system to become an MCP Server and made an MCP Client from my ToolAgent. I use zero dependencies on other frameworks, just to see how the protocol works.
starkravingfinkle.org/blog/2025/03...

31.03.2025 15:15 โ€” ๐Ÿ‘ 0    ๐Ÿ” 0    ๐Ÿ’ฌ 0    ๐Ÿ“Œ 0

This is pretty interesting. After building something similar (WebAgent), I can understand what the code is doing. I also exposed Playwright as "tools", but I didn't realize Playwright has a a11y snapshot feature to get a reduced set of elements. Very handy.
starkravingfinkle.org/blog/2025/03...

25.03.2025 01:54 โ€” ๐Ÿ‘ 0    ๐Ÿ” 0    ๐Ÿ’ฌ 0    ๐Ÿ“Œ 0
Exploring LLMs as Agents: WebAgent Gets a Refactor My fifth post on LLMs as Agents. I just itโ€™s a series now. After refactoring ToolAgent, I decided to do the same with WebAgent, building it on the ToolAgent system. For more context and a refโ€ฆ

With the new ToolAgent framework, I refactored my web automation agent into WebAgent. The Playwright-based automation actions are now tools used by the agent.

I had to stop sending the entire HTML content of the page to the LLM ๐Ÿ˜ฑ
starkravingfinkle.org/blog/2025/03...

20.03.2025 00:26 โ€” ๐Ÿ‘ 1    ๐Ÿ” 0    ๐Ÿ’ฌ 0    ๐Ÿ“Œ 0
Exploring LLMs as Agents: Tools & Benchmarking I spent some time refactoring the Tool Agent code, added some additional mock tools and even some basic benchmarking. For more context, checkout the previous posts: Minimalist Approach: I kicked ofโ€ฆ

I'm still exploring ("playing") with using LLMs as Agents. I refactored the code into a tiny framework. Have I become that which I wanted to avoid? Maybe ๐Ÿซ 

More tools and a benchmark system allowed me to really improve the agent success outcomes.
starkravingfinkle.org/blog/2025/03...

16.03.2025 16:55 โ€” ๐Ÿ‘ 1    ๐Ÿ” 0    ๐Ÿ’ฌ 0    ๐Ÿ“Œ 0

Totally agree. I hope "it improves testing, reduces complexity, allows us to move faster"are also true, or you're likely better off doing the simplest thing possible.

16.03.2025 16:50 โ€” ๐Ÿ‘ 1    ๐Ÿ” 0    ๐Ÿ’ฌ 0    ๐Ÿ“Œ 0
โ€œStranding all or part of rural America with worse internet so that we can make the world's richest man even richer is yet another in a long line of betrayals by Washington,โ€ said former BEAD director Evan Feinman

โ€œStranding all or part of rural America with worse internet so that we can make the world's richest man even richer is yet another in a long line of betrayals by Washington,โ€ said former BEAD director Evan Feinman

BREAKING: The head of the $42 billion federal BEAD program that aims to bring broadband internet to rural and unserved areas has left the Commerce Dept.

He sent a departing email warning of "deeply negative outcomes" for people if the Trump admin shifts $$ from fiber internet to Musk's Starlink:

16.03.2025 15:45 โ€” ๐Ÿ‘ 4892    ๐Ÿ” 2185    ๐Ÿ’ฌ 96    ๐Ÿ“Œ 182

@mfinkle is following 20 prominent accounts