Head of R&D at MiniMax Code. Co-creator of DeerFlow. Building the next generation of AI agent systems.
北京
Joined March 2025
- Tweets718
- Following273
- Followers7.9K
- Likes777
RonnyHe retweeted
Introducing Claude Fable 5.2 and Claude Mythos 5.2, designed to tackle the most daunting technical challenges.
They’re capable collaborators on ambitious projects in many fields, and show improvements in areas like coding, reasoning, and vision.
RonnyHe retweeted
A lot of content engineering challenges have pretty standard solutions, which has lead to some engineers saying agent engineering is solved/boring
True to some extent, but there are a bunch of (fun) new ones emerging as agents evolve!
RonnyHe retweeted
we will need to rebuild and rewrite much of software to adapt to a new, faster pace.
code used to arrive one drop at a time. soon, if not already, code will be written as if it were a continuous stream, like a violent river twisting and turning as it pleases.
this is not a new thing: big tech cos have had to deal with this with the enormous scale of their engineering teams. but now ai has brought an incredibly elastic and capable workforce wielded by every engineer. this is your problem now, and the only way to adapt is to evolve your systems so agents - even and especially the ones with little context about the right way of doing things - can thrive. make the easy thing the right thing.
your codebase is becoming organic architecture. agent contributions grow around the constraints and guardrails you’ve designed. good software engineering matters even more now! and no, the continuous stream of code is not something you have to watch helplessly. be more ambitious and design systems, frameworks, and languages to control the chaos.
the time for mourning coding is long over
RonnyHe retweeted
Totally agree! GPT models especially love coming up with edge-case race conditions. Older Claude versions did this too. This prompt is definitely worth trying.
This is such a banger rule in @opencode's AGENTS.md
"Before adding complexity for a speculative or vanishingly unlikely race or security edge case, explain the concrete failure mode, likelihood, and complexity cost to the user and get their buy-in. Do not silently expand scope for theoretical robustness."
RonnyHe retweeted
I put Anthropic's whole harness playbook for long-running coding agents on one page
honestly the most useful thing I've read on agents all year
two Anthropic engineering posts + the Opus 5.5 launch, squeezed into one sheet:
> a planner grows a one-line idea into a 16-feature spec
> a generator builds it, an evaluator opens the app in Playwright and fails any sprint below the bar
> session one writes init.sh, a progress log and 200+ failing features. every session after closes one and commits
> the 4 ways agents wreck long tasks, and the fix for each
> 3 prompts to copy: coding agent, QA agent, planner
solo run: 20 minutes, $9, the game didn't even respond to input
full harness: 6 hours, $200, playable
QA was about 8% of the bill on their DAW run
and Opus 5.5 runs about 40% cheaper than Opus 5
bookmark this before you wire up your next agent
RonnyHe retweeted
Two offers, now live in MiniMax Code:
- Unlimited M3.1 Flash
Through Oct 25, 23:59 (UTC+8), every MiniMax Code subscriber can use M3.1 Flash with no usage caps.
- 50% off your first month of M Plan
Through Oct 31, 23:59 (UTC+8), new subscribers and M Plan upgraders get 50% off their first month. Standard pricing applies from month two.
RonnyHe retweeted
Writing code is easy now. You can build an app just by saying what you want. But things are very different in production. I'm a hands-on developer, and writing code may account for less than 10% of my total delivery time. Most of the rest goes into review and verification.
No matter how far models advance, I never seem able to fully let go. I always have to run their code myself to see whether it meets the requirements correctly. This is the main bottleneck keeping me from delivering at scale.
So the way I see it, the key to large-scale parallel delivery is letting models verify the code they write, and making that verification something we can trust.
If I could keep only one of the Skills I've written over these two months, I'd keep the one that builds the verification environment: the agent starts the app itself, uses it like a user, then checks the results against a checklist organized by feature, with an ID for every criterion. I call it a feature map.
How I built it and the problems I ran into are all in this article.
RonnyHe retweeted
A Fresh New Look
MiniMax Code has a refreshed brand identity and a redesigned interface, along with subtle interaction refinements that make everyday use smoother.
Upgraded Agent Capabilities
MiniMax M3.1-Flash-Preview makes its debut in MiniMax Code. This update also brings improvements to Computer Use, MiniApp, custom themes, the Git Graph panel, real-time agent action tracking, and more.
M Plan Coming Soon
M Plan will succeed Token Plan and expand its subscription offerings. Subscription pricing and text usage quotas will remain unchanged, and select tiers will unlock the full MiniMax model lineup, including H3.
Token Plan will no longer be available for new purchases. Current subscribers with auto-renewal turned on will keep their existing plan unchanged, and they can upgrade to M Plan.
*These offers and features will be available after the MiniMax Code v3.1.0 release.
RonnyHe retweeted
🚨🚨🚨: New Release
Must have enhancement Pack for the MiniMax Code/CLI (MCode) internet browser support, JEV integration, playwright integration, Context compaction fix, scroll bar for console transcript, and context progression bar. LFG!!!!..
Full GItHub Repo:
github.com/drowzeys/keys-Min…
