Ally Weir

@allyjweir.co.uk

Senior Software Engineer working on system reliability and availability. Interested in Systems Thinking and Human Factors.

I've found reading proposals and one-pagers difficult more recently as AI tells are harder to spot. I can't tell if the author is thorough and deep in the context of the problem or if they churned it through a model for 10 minutes. The only way to tell is get on a call and just test knowledge.

An internal discussion post at work reminded me of the "Celebrity Timeline" problem that Twitter had to re-architect around back in ~2013. Revisited this post about how Bluesky solves the problem today with "lossy timeline" being an acceptable trade-off for the domain. jazco.dev/2025/02/19/i...

When Imperfect Systems are Good, Actually: Bluesky’s Lossy Timelines

By examining the limits of reasonable user behavior and embracing imperfection for users who go beyond it, we can continue to provide service that meets the expectations of users without sacrificing s...

jazco.dev

My most recent side project met reality in the past few days as it was rolled out to its first real users. Lots of learnings about the experience. My agent-assisted dev-flow has evolved quite a bit since I first built it out.

GPT-5.6 Terra has converted me. It's been a really great model for my use at work. Before I had been a Claude loyal but Terra has stayed on the rails and done good work when I needed it. I've also been leaning into "Auto-mode" in Copilot for many tasks. It's been way better than before.

atproto is super interesting and would be something I'd love to poke at if I had more free time. In particular it'd be really cool as a pet project to learn Clickhouse a little more in-depth than I do just now.

Scoiattolo@scoiattolo.mountainherder.xyz · 3w ago

One thing I like to do is to check clickhouse to see what lexicons have been the most active over the last two months. That's how you learn that a Japanese study app (studynext.app) is the second most active Atproto application right now.

A markdown table showing the current top 25 top-level NSIDs: 

A dark-themed table ranking namespace ID roots by activity. The columns are “#,” “nsid root,” “events,” “distinct lexicons,” and “share.”

Visible rows:

| #  | nsid root                  |      events | distinct lexicons |  share |
| -- | -------------------------- | ----------: | ----------------: | -----: |
| 1  | app.bsky                   | 338,480,884 |                24 | 96.29% |
| 2  | app.studynext              |   2,156,118 |                23 |  0.61% |
| 3  | cx.vmx                     |   1,151,099 |                 2 |  0.33% |
| 4  | site.standard              |   1,113,465 |                 5 |  0.32% |
| 11 | link.bridgebeats           |     392,316 |                 1 |  0.11% |
| 14 | community.lexicon          |     245,161 |                16 |  0.07% |
| 15 | games.gamesgamesgamesgames |     244,932 |                11 |  0.07% |
| 16 | team.sigint                |     156,514 |                 3 |  0.04% |
| 17 | net.anisota                |     131,988 |                34 |  0.04% |
| 18 | org.titlegraph             |     105,841 |                 6 |  0.03% |
| 19 | is.currents                |      96,942 |                 5 |  0.03% |
| 20 | com.nsc                    |      96,855 |                 1 |  0.03% |
| 21 | org.hypercerts             |      92,294 |                20 |  0.03% |
| 22 | ai.afrilingua              |      86,710 |                15 |  0.02% |
| 23 | com.puzzmo                 |      83,180 |                 3 |  0.02% |
| 24 | jp.5leaf                   |      66,824 |                 1 |  0.02% |
| 25 | app.qte                    |      53,658 |                14 |  0.02% |

First game on the Lenovo Legion Go has been Prey: Mooncrash. Wow, it is so close to Deathloop. Amazing to see the origins on that game. Deathloop is underappreciated, not perfect but interesting for sure. Even more so with Marathon now.

After much deliberation, I've finally ordered a handheld. I saw a good deal for the Lenovo Legion Go S: - has SteamOS, receives updates - slightly more powerful than Deck - more compact - was cheap In the end I think I was over-analysing a bunch. Just needs to play Steam games.

Of all Xbox's numerous gaffes over the past decade+, how they've handled Arkane Studios hurts the most. Produced a run of incredible games from Dishonoured through to Prey and Deathloop was a solid if not perfect entry Since then they haven't been able to thrive.

Heard the distinction between "agent-driven" and "agent-assisted" development and how different models can be better at each of these. Personally I've found recent Opus models running wild on their solutioning. Multiple rounds chewing through tokens to create workarounds to non-existent constraints

Early LLM-powered software engineering practices will be summarised to: Just because you can, doesn't mean you should. I'm looking forward to seeing what the future of this is beyond the current milieu

A frustration with AI-infused development is when folks aren't using it to solve problems at hand but instead add extras that aren't necessary nor particularly value-add. Just because it's cheap doesn't mean we should be doing it.

LLMs are not an opportunity to abdicate expertise or build your personal knowledge. Seeing folks stop growing or learning because 'the machine can do it' is heartbreaking quite honestly. The folks I've enjoyed working with most have been those focused on adapting and augmenting, not replacing.

“the economic order must remain subordinate to human dignity and the common good” I have news for some folks. The pope is a socialist.

It's killing me how difficult it is to get a match in Marathon. Every time I've tried recently it has been 10+ minutes then the match lasts less time than that. Had a lot of fun those first couple weekends with it. Hoping the season reset will bring it back to playable for me.

Had a real experience of EV infrastructure in Northern Scotland this weekend. Broken chargers, 7kW the fastest available and 87p/kWh prices! Then Ionity chargers makes things all the better. If only there were more of them!

Tried for 30 minutes last night to get a match in Marathon. Unsuccessful across different maps, queue types etc. When I can get an Arc Raiders match in <2 mins, waiting 10 and not getting a bite is pretty frustrating. I know where I'll go when I've only got limited time.

It's been really frustrating to see VSCode's performance slip in the past 12 months. I liked it for start-up speed, responsiveness and that's just not the case anymore. Even on my fast machine. Glad we've got a solid ecosystem of alternatives like Zed to reach for.

Coming from the 'software as craft' camp of the industry, the LLM-as-slots-machine model of software development is trying. It tilts conversations with other contributors massively. Do they back their contribution? Would they be upset if I changed what they'd helped shape up with the slots?

So many tiny design details are frustrating with Word Online compared to GDocs. From not opening docs in new tabs, unable to move between individual words and the share tools being really poor. Don't sleep on just how good GSuite actually is.