Pete Hodgson (@thepete.net on bluesky)

7.6K posts

Pete Hodgson (@thepete.net on bluesky) banner
Pete Hodgson (@thepete.net on bluesky)

Pete Hodgson (@thepete.net on bluesky)

@ph1

@thepete.net on bluesky Independent consultant helping engineering teams tackle thorny problems. Sociotechnical architect 🧐. Formerly Earnest, ThoughtWorks.

Pacific Northwest, USA Katılım Mart 2009
468 Takip Edilen3.2K Takipçiler
Pete Hodgson (@thepete.net on bluesky) retweetledi
Matt Pocock
Matt Pocock@mattpocockuk·
Doing some experiments today with Opus 4.6's 1M context window. Trying to push coding sessions deep into what I would consider the 'dumb zone' of SOTA models: >100K tokens. The drop-off in quality is really noticeable. Dumber decisions, worse code, worse instruction-following. Don't treat 1M context window any differently. It's still 100K of smart, and 900K of dumb.
English
151
58
1.2K
149.8K
Pete Hodgson (@thepete.net on bluesky) retweetledi
boris
boris@boristane·
slop creep is what happens when you turn your brain off and hand the thinking to coding agents each individual change is fine, but all together, you have a pile of crap we're witnessing this happen in real-time across everything boristane.com/blog/slop-cree…
English
41
63
653
88.3K
Pete Hodgson (@thepete.net on bluesky)
Being an Old, I have a bit of nostalgia for The Good Old Days of OSS where you shared a thing and maybe some people used it, and there wasn't any influencing or fancy websites or weird drama. It's nice to rediscover that vibe in the 3D printing community...
Pete Hodgson (@thepete.net on bluesky) tweet media
English
1
0
2
134
Sam Parr
Sam Parr@thesamparr·
How is everyone getting team adoption for Claude? I spent a lot of time on Twitter, as do you. We see all this AI stuff popping up. We're on top of it, or at least sorta. I know what's going on and are testing all these fringe ideas. But how are all you people getting your team to actually use it effectively without spending all their time on Twitter and learning, which we know they won't and probably shouldn't be?
English
274
21
513
282.9K
Pete Hodgson (@thepete.net on bluesky)
@0xblacklight Amazing write-up! Can I steal your subagent context window visualization for a presentation (w. credit!)? Also FYI in "Distributing Tools with Skills" you say you can't package MCPs, scripts etc. in a skill. It's true, but Claude Code's plugins solve exactly for that.
English
1
0
1
94
Pete Hodgson (@thepete.net on bluesky) retweetledi
dex
dex@dexhorthy·
Here’s what’s gonna happen: - you replace your code review with feedback loops (sentry, datadog, support tickets, etc) - you stop reading the code - software factory fixes everything - one day something breaks at 3am, agent can’t fix it - nobody’s read the code in 3 months - you have 3 weeks of downtime trying to re-onboard and fix it - you lose significant % of your contracts and users - your company is now dead
dex@dexhorthy

@gregpr07 this may surprise you that thus is coming from me but I think we’re in for a 1-3 year period where stuff might break at 3am and if you’re relying on loops to fix it and nobody understands what’s under the hood, you’re looking at an existential threat to your company

English
258
567
6.9K
588K
Pete Hodgson (@thepete.net on bluesky) retweetledi
dax
dax@thdxr·
sent this to the team today everything great comes from being able to delay gratification for as long as possible and it feels like we're collectively losing our ability to do that
dax tweet media
English
254
706
6.9K
960.2K
Gergely Orosz
Gergely Orosz@GergelyOrosz·
Exactly one year ago (10 mar 2025), Dario Amodei: "I think we will be there in 3-6 months, where AI is writing 90% of the code. And then, in 12 months, we may be in a world where AI is writing essentially all of the code." This turned out to be... too darn accurate.
English
239
168
3.8K
453.8K
dex
dex@dexhorthy·
Yes harness used to be called agent but we can’t have nice things
dex@dexhorthy

@leo_trapani Yeah well that used to be called “agent” but the word “harness” was made necessary because the saas slop industrial complex broke our terminology by trying to call everything an “agent”

English
3
0
13
3.2K
dex
dex@dexhorthy·
can we plz settle on what a "harness" is quickly harness == the coding agent (claude code, codex, etc) apparatus == the stuff you build around the coding agent (backpressure, mcp, ralph wiggum, etc etc) i'm tired of this blurry line where ppl use "harness" to mean both the agent and the stuff you build around it
English
95
14
284
35.2K
Pete Hodgson (@thepete.net on bluesky)
@somi_ai @dexhorthy Even better is to write a custom lint rule that checks for await with cookies, and make sure the agent lints after every edit. There are lots of more nuanced guardrails that can't be expressed just via static analysis, but where you can it's a huge win IMO
English
0
0
0
45
Somi AI
Somi AI@somi_ai·
@dexhorthy yeah we found this too. the fix was keeping CLAUDE.md under 200 lines and being brutally specific. vague rules like 'follow best practices' get ignored, but 'always use await with cookies()' gets followed every time
English
3
1
59
5.3K
Peter Steinberger 🦞
Peter Steinberger 🦞@steipete·
Been wrangling a lot of time how to deal with the onslaught of PRs, none of the solutions that are out there seem made for our scale. I spun up 50 codex in parallel, let them analyze the PR and generate a JSON report with various signals, comparing with vision, intent (much higher signal than any of the text), risk and various other signals. Then I can ingest all reports into one session and run AI queries/de-dupe/auto-close/merge as needed on it. Same for Issues. P rompt R equests really are just issues with additional metadata. Don't even need a vector db. Was thinking way too complex for a while. There's like 8 PRs for auto-update in the last 2 days alone (still need to ingest 3k PRs, only have 1k so far).
Peter Steinberger 🦞 tweet media
English
425
211
4.1K
568.4K
Angie Jones
Angie Jones@techgirl1908·
Developers who are full agent mode: If all of this went away tomorrow, do you think you could go back to writing code by hand? Like would you even want to?
English
197
11
276
56.6K
Angie Jones
Angie Jones@techgirl1908·
@sudo_overflow no like they will just remove an assertion or delete the whole test if it fails
English
3
0
1
221
Jarred Sumner
Jarred Sumner@jarredsumner·
In the next version of Claude Code Claude Code’s Bash tool runs faster and uses less memory. Claude writing 1 GB to stdout 10 times: Before: 115 seconds After: 15 seconds (7x faster)
Jarred Sumner tweet media
English
166
155
4.3K
797.4K
Pete Hodgson (@thepete.net on bluesky)
It's like running shoes. For serious athletes there's definitely a difference between your ideal shoe and a mediocre shoe. But for MOST OF US, we just need to lace up something non-terrible and get some miles in.
David Cramer@zeeg

tbqh the whole industry should realize it its just an argument of vim vs emacs use a top N harness+model, and then focus on how to use the technology vs pretending a fractional improvement to the model is going to unlock things for you

English
0
0
1
223
Pete Hodgson (@thepete.net on bluesky)
When the AI-pilled startup CTO brags "none of our engineers have opened an IDE in months", I'm left wondering: How are y'all reviewing all that AI-written code?! In the github UI, like a savage? In a TUI? No need to review it, cos "it's just assembly language now"?
English
0
0
2
272