Max Shaw
835 posts

Max Shaw
@_maxdshaw
Building @trywindmill. Built @GifJifApp. Formerly Product @Yext,
New York, NY Katılım Şubat 2010
813 Takip Edilen458 Takipçiler

What kind of IC work can you do if you are a manager?
Attended two dinners with dozens of senior Bay Area tech managers last week.
Every single one of them said they were expected to be hands-on building with AI these days. But what to work on?
4 categories emerged:
1. Internal efficiency -- anything to make your team more productive, from communication tools to digital brain to team skills.
2. Quality-of-life improvement -- got something that bugs the heck out of you about the product but you can't convince anyone else to work on it? Just fix it yourself.
3. Celebration story -- artifact (usually video, image narrative, etc) that hypes up the work done by your team.
4. Vision piece -- describes a really freakin' cool future that your team can move towards.
What NOT to do: take on any critical-path product work. (Because then you either do a poor job delivering on that, or managing.)
English

FYI, all my plans on these dates are hereby cancelled.
Steve Popper@StevePopper
NBA makes start date official.
English

@chrispisarski How are you talking to this c level person in the first place if you havent gotten permission
English

stop asking prospects for a "quick 15 minute call"
if your prospect is a c-level or vp, 15 minutes of their time is incredibly expensive. they literally won't give it to you just because you asked nicely
during YC, we stopped asking for time and started asking for permission to send information:
"hey mark, i analyzed how your sales team is doing outbound and noticed you have a huge bottleneck in step 2. i recorded a 2-minute video showing exactly how to fix it. opposed to me sending it over?"
people will happily say "sure, send it". once they watch the video and realize you actually know what you are talking about, they will ask you for a meeting
English

@brian_lovin Why aren’t people talking about this more! It’s so frustrating.
English

Yeah makes sense! Thanks for the write up.
Harnesses definitely matter but I just don't see them as this huge strategic moat. And yes i constantly switch between different ones depending on the task. The UI around the harness (worktress, browsers etc) also matter a ton.
One example - I don't think gemini has failed because they don't have a good harness. Curious if you agree with this take from the legend @benthompson or not.

English

LOL see latest tweet as i automate myself away
@_maxdshaw - you're not missing a bunch depending on your tasks but harnesses with today's model intelligence has a pretty large impact on Task specific perf (see some of our LangChain blogs in my highlights, Cursor blogs, OpenAI prompt guide per model, Terminal Bench 2.0 empirical results, Factory blogs)
part of this is that models today are post-trained in certain harnesses so they get better at reasoning, tool calling, etc given "similar" tool-shapes and prompts
but actually I think that fit matters less than the Task in Model-Harness-Task fit --> different tasks require different prompting, tools, skills, problem decomposition
for hard tasks telling the model to "just do it" doesn't work today as much as I wish it would, so the harness helps guide the model towards solving the Task by injecting human+agent design priors
often these take the shape of context management on behalf of models, bundled skills, and an ability to route to specialized models for certain tasks
i hope this helps! tldr: maybe not missing anything but today it's worth leaning into imo, maybe this becomes less important in the future, but might as well cook on it a bit now
English

@ThePrimeagen 20 years ago: Software engineering is dead. There's no manually written assembly anywhere.
English

SaaS you buy for yourself/personal use - AI gonna kill it
SaaS you buy for a team - not so much
@levelsio@levelsio
I replaced almost all my SaaS subscriptions with my own vibe coded replacements But n=1
English

@mignano Totally agree. Memories and models are transferable but the harness isn’t. Anthropic seems to know it. My API token usage costs ~10x what I pay for the same work through their flat-rate harness plan. They’re clearly subsidizing harness usage to lock people in.
English

@typesfast Does anyone have an updated version of this chart? Curious with the latest data
English


@RhysSullivan Does this need to get merged in order to use it? Do I need to fork portless?
English

@nateberkopec @stratechery He’s uncharacteristically off when it comes to harnesses. Wonder where his intel is coming from.
English

It's very weird to hear @stratechery constantly opine that "models and harnesses are tightly integrated" when no one I know of on X thinks that, and usually thinks the opposite (Claude performs better in non-Claude Code harnesses)
English

We added 10+ AI agents into Slack as AI employees and now it just feels like information overflow.
Too many messages, hard to tell what actually matters.
Not sure chat is the right interface for this tbh.
How are people dealing with it? voice? visuals?
Maybe I’m biased, but I keep wanting something more visual. Feels like you can process way more, way faster.
English

@thesamparr My order is a little nutty… lox, whitefish, scallion cc, capers, onions on an everything bagel
My options are limited :(
English












