Avery Jones

tokens don't grow on trees.
context engineering.
made

[Avery Jones @Avery_Jones]

@dexhorthy agent docs are a shortcut agents can take to navigate more efficiently and consume denser knowledge

yes stale docs are worse than no docs, but in the age of self-maintenance, why not include them?

[Avery Jones @Avery_Jones]

@matanSF @FactoryAI super cool approach to the tool calls! crazy efficiency gains to go from 3M tokens -> 60k

[Avery Jones @Avery_Jones]

hacktober takes more victims

[The Kobeissi Letter @KobeissiLetter]

BREAKING: Binance has been hacked with wallets across the Ethereum, Bitcoin, and TRON networks seeing outflows of -$5.7 billion.

The hack includes $1.45 billion worth of Bitcoin and $875 million worth of ETH.

If confirmed, this would mark the biggest hack in crypto history.

We expect more information shortly.

[Avery Jones @Avery_Jones]

@sydneyrunkle @Stagehanddev Very cool! Are they able to do screenshots etc?

[Avery Jones @Avery_Jones]

@mattpocockuk wow, thanks for sharing. crazy default to set

[Avery Jones @Avery_Jones]

@theo That’s be awesome, not fun having to manage the underlying providers individually

[Avery Jones @Avery_Jones]

agentic atlas this week. first thing I've made that people actually use, still kinda can't believe it!

[Avery Jones @Avery_Jones]

the checks I trust on agent work are the boring ones. a test or a hook gives the same answer every time. a second model grading the first is just another opinion

[Avery Jones @Avery_Jones]

@strimblez makes sense. I’ve been focused on what the reviewer catches, but i should be looking at what it misses too. i like the idea of sampling passes and tracking it by change type, will give it a try. thanks!

[Avery Jones @Avery_Jones]

opening the vault on some of my skills that help me ship fast without supervision.

https://t.co/MdarkBEHte

[Avery Jones @Avery_Jones]

thanks for sharing! one bit of context: i’m running this headlessly, so i’m not reviewing every PR myself. the hard part for me is knowing when i can trust the agent’s “done” without being in the loop. even with tests, my reviewer has caught cases where removing a spec’d guard leaves the suite green. curious how you think about building confidence in that kind of autonomous workflow?

[Avery Jones @Avery_Jones]

@abrilzucchi 2/4 ain’t bad

[Avery Jones @Avery_Jones]

@hmartenjoyer @typesafeai need the bird

[Avery Jones @Avery_Jones]

@strimblez the earliest tell for me is untested guards. it writes the check the spec asks for, but nothing fails if you delete it. my reviewer agent mutation-tests every PR and ~a third of blocking findings are exactly this. next step is the PR description claiming more than the code does

[Avery Jones @Avery_Jones]

MCPs are a pain to add - ideally they can be added on demand or limited to an agent type

Usually use https://t.co/aJ2TzJafVI since it’s so much easier to add and manage, but doesnt provide the full plugin

Marketplace and plugin separate updates causes version issues

Local dev of a marketplace and switching between live and dev versions is a mess

[Avery Jones @Avery_Jones]

@Polymarket pace the frontier

[Avery Jones @Avery_Jones]

@sydneyrunkle What is it about human messages that doesn’t work? Just differentiation in authority / source?

[Avery Jones @Avery_Jones]

@weswinder I already am a software factory. Just formalizing it

[Avery Jones @Avery_Jones]

@shiraeis totally agree! I feel like I woke up in a childhood dream where I can imagine things and they become real

[Avery Jones @Avery_Jones]

@ahmetb What’s another one on the pile?