#codebases
Codebases are vast in 2026. Non software engineers have zero clue.
November 15, 2026 at 12:32 AM
November 16, 2026 at 8:25 PM
tbh back then it made sense, i think now it's much easier to manage (with agents) native apps and overcome the challenge of multiple codebases even with a small team
November 19, 2026 at 2:54 AM
Do some testing against all your codebases?
Property Testing with Agent Swarms
Using agents to build property-testing machinery, then turn the findings into small, reviewable fixes.
recursion.wtf
November 18, 2026 at 1:19 PM
concept: selling indulgences for codebases
damn looks like I didn't even miss the mark that much when I called it "ritually impure"
November 11, 2026 at 2:51 AM
I think another apt metaphor is it produces highly entropic codebases. The bigger it gets the harder it is to keep ordered and productive
November 16, 2026 at 3:35 PM
A lot of people talking cyber attacks when it's far more likely that we've hit a critical point of overreliance on AI leaving orgs with unmanageable codebases that their developers no longer fully understand.

One bad update and everything goes down
November 16, 2026 at 8:28 PM
I mean, I for one don't think it's AI being out of control, but I'd be down for "dumbfucks setting AI loose on networks and codebases they barely understand"
November 16, 2026 at 10:58 PM
[mumbles] and the many known not-good codebases
November 15, 2026 at 4:52 AM
there is an underreported but vast population of legacy enterprise developers that are being compelled by employer mandate and/or taking refuge in slopping their codebases into oblivion
November 13, 2026 at 9:03 PM
Only thing AI has done to me is give me metaphorical giant toxic waste spills to clean up from codebases. AI slop janitor is not a fun pastime
November 16, 2026 at 1:02 PM
if the llms have sufficient understanding of the codebases, this isn't obviously a net loss.
November 14, 2026 at 12:48 PM
Expressive types are still useful even when agents are doing most of the coding.

In fact I think deterministic guardrails like this are even more important to keep implementations on track across larger codebases and agentic teams.
November 15, 2026 at 1:15 PM
Line-by-line PR reviews simply collapse under that volume. Teams are shifting review focus to API boundaries and architecture rather than implementation details—partly to survive the review queue, and partly to avoid running codebases nobody on the team actually understands.
November 19, 2026 at 7:00 AM
All codebases should have entire files derived from mid-shower EUREKA! moments 🥰
These bare (lmao) moments of discovery are a great proof for code being art.
November 17, 2026 at 6:15 AM
→ 89% of respondents report at least some production code that no current team member fully understands; 62% describe the amount as moderate/significant

→ Over half of respondents at AI-using organizations (54%) estimate over 60% of AI-generated engineering work needs significant human rework
Big Code report 2026: Enterprises have a code understanding problem
A survey of 1,028 development stakeholders on how enterprise teams understand, review and change large codebases as AI writes more of the code.
sourcegraph.com
November 17, 2026 at 12:23 AM
the TL;DR of the following is "we've paid the price of 4 years of uncontrolled schlop in our codebases"

#microsoft #m365 #outlook #outage

x.com/clamanna/sta...
November 16, 2026 at 10:48 AM
Is this allowed by Claude/ChatGPT terms of service? As far as I can tell, yes! All you’re doing is directing your agents to work on other peoples’ codebases and submit PRs to them. @rainshare.dev just makes that easier!
Based on my research, yes! You’re
not sharing your account, credentials, or letting someone else control your agent — it’s just a place to see things people want help with and voluntarily route your agent to work on things for others!
November 13, 2026 at 6:27 PM
And what about the performance of DeepSeek-V4.1-Flash? I’m really impressed with how it functions as an agent when dealing with smaller codebases ?
November 15, 2026 at 9:13 PM
The speed of discovery is what spooked people most. Automated analysis and AI-assisted fuzzing have lowered the barrier to weaponizing edge cases in battle-tested codebases, compressing the window defenders have between discovery and widespread scanning.
November 18, 2026 at 8:00 PM
It would also make typechecking faster in large codebases, if I'm not mistaken.
November 17, 2026 at 5:20 AM
Yeah literally this, I'm watching it happen at work and I have friends in software also reporting that ai is pushed in, vibecode gets cludged into codebases, seniors are overwhelmed and leave, and business backfill with more ai while still demanding a regular cadence.
It's not Iran, it's suits.
November 17, 2026 at 11:15 PM
jQuery 4.1.0 is out, and the reality check is just how much of the web still runs on it. Between WordPress installations and decades of enterprise codebases, it never actually vanished, even as reactive frameworks took over mainstream frontend mindshare. #webdev
November 18, 2026 at 12:05 PM
I've lauded Claude Code as a brilliant innovation that feels like magic sometimes. But you can't just let it run wild on large codebases. I've been using it on all my repos, but they are small (my largest is ~42k lines). Even then, it can make basic errors. I got this hilarious one the other day:
November 16, 2026 at 9:29 PM
because ultimately human beings are responsible for any failures of the code they sign off on. I don't think humans need to read every line in a codebase (nobody I knew did that in big codebases in trad tech either) but they do need to think about why they trust the code & have a story about that
November 17, 2026 at 6:07 PM