Ally Weir

In light of this week’s PHP story, it feels worthwhile sharing this article sheets.works/data-viz/hol...

Tool exhaustion related to AI is something I'm experiencing and speaking to others they are too. There's so many different surfaces, tools, workflows. That then creates duplicate work to get set up and ends up with a fragmented experience across the different surfaces.

A single surface that allows for: - durable identity agents over time that I can assign tasks - ability to schedule regular background work - sandboxed - network policy controlled - unique auth identity, safely mediated Beyond this, I want to see how we get the social aspect of AI workflows solved

Tool exhaustion related to AI is something I'm experiencing and speaking to others they are too. There's so many different surfaces, tools, workflows. That then creates duplicate work to get set up and ends up with a fragmented experience across the different surfaces.

To try and help, I've really leaned back into custom agents and skills in one big directory. Just using the Copilot CLI for everything that I can. Narrowing back, focusing on how to make that experience better rather than reaching for a completely new conceptual model for each new goal I have.

Really cool to see some of the work coming out around agent runtimes. The early consolidation around OpenAI's chat and completion API model was right then but there's a whole layer ("substrate" apparently) of common functionality that is now being abstracted.

Poking a little around the edges of these efforts, it's cool to see Kubernetes folks building around this. And Google open sourcing (one of) their versions, Ax. agentexecutor.io

Hackathon over the weekend on a new OSS project, working with friends at other companies. Hadn't realised how much I took `/yolo` plus sandboxed dev environments for granted. Clicking 'Allow' on every prompt in VSCode with yolo locked behind an admin setting. 😬

AI has made it difficult for teams to estimate "time to deliver" or sizing effort. When AI works, you can go lightning fast. But what I've seen is project scope creep like never before. Simple efforts become defining new frameworks and building net-new toolchains.

Yes, I sound like that grumpy senior you worked with early in your career. As we rocket through this chaotic era, balancing innovation with delivery is more important than ever.

AI has made it difficult for teams to estimate "time to deliver" or sizing effort. When AI works, you can go lightning fast. But what I've seen is project scope creep like never before. Simple efforts become defining new frameworks and building net-new toolchains.

Instead, I come back to the concept of "innovation tokens" in projects and in tasks I'm completing. We should limit innovation tokens and spend them where they will make the most difference. What is our project's unique value? Let's spend those tokens where we can make that shine.

I've yet to dip into any of these agent platforms like OpenClaw yet. Definitely been curious about projects like Buzz and, to an extent, Grok Bot. But also, the Grok ick has touched this one. Makes me think it's worth waiting a few more months once these ideas have made it into other projects

I dug into this more. Key innovation for me is durable agentic personas that can be addressed and act as contributors/collaborators rather than the work of AI being obscured behind a human's Personal Access Token. I want to know when the human did it vs. the AI.

Inspired to finally read @charity.wtf's post this morning after a colleague told me they had started receiving DMs written entirely by AI in the past few weeks. charity.wtf/p/confession...

I come to work to learn from others, collaborate and build systems that help make people's lives better. AI can help us do this better but fuck me, when we're replacing the human connection with it what are we doing. Spending time with others and creating connection with others is what matters.

Picking the right tools can save literally years of development. The CNCF is such a treasure trove of incredible technology just sitting there waiting to be used, contributed to and shared. Teams sometimes don't realise how good we've got it when it comes to OSS

I've yet to dip into any of these agent platforms like OpenClaw yet. Definitely been curious about projects like Buzz and, to an extent, Grok Bot. But also, the Grok ick has touched this one. Makes me think it's worth waiting a few more months once these ideas have made it into other projects

Building a new product from scratch can be a real challenge at any time, now with AI it's even more challenging. When anything is possible, and you can brute force solutions with tokens, how do you help a team remain focused on seeking the real value and delivering on that.

Telling a convincing story is certainly part of that. Building narrative that folks _want_ to buy into. A challenge I've faced is the greenfield project slipping towards the "everything app". It should solve all problems for all people in any way. And be ready in the next month!