Instead of having this moltclawd thing installed on a Mac and posting on reddit it'd be more interesting to have the agent installed on a cloud Linux instance, give it the ability to submit code, push code to production, and spin up other agents in their own Linux instance.
See if they start expanding their own capabilities by deploying their own tools and spawning other agents recursively.
This whole Clawdbot thing reminds me of the early consumer file and photo cloud storage days.
There were countless startups and DIY solutions out there. Yet, at the end, the winners were the platform's built-in solutions (e.g. Google Drive, Google Photos, iCloud, Apple's Photos).
Dropbox came the closest to be a breakout success in the consumer storage space yet they had to shift to Enterprise since the platform's built-in solutions became good enough. Like Steve Jobs said, they were a "feature, not a product".
The AI space is playing out in a similar way.
It's likely just a matter of time before the platforms themselves build these features in.
It's not trivial for startups to do this because they don't have the models nor the user data, they're just acting like glue (hence the MCP trend that went nowhere).
There is still room for Dropbox and Box-like startup successes but it's harder this time given the above. Anthropic and OpenAI would be the likely contenders IMO.
There's also room for alternate clients like Superhuman but they're working with bigger constraints and are mainly a bet on being able to provider better UX than the platform they're building on top.
I don't know much about Enterprise but there are likely strong opportunities there as well. Although players like Microsoft, Salesforce and GCP are likely making headway there.
Is anyone actively using Clawdbot?
I'm just curious whether anyone finds it actually useful and for what purpose.
I've scrolled through their "showcase" feed and none of it seems useful to me.
Last Friday was my last day working on Farcaster.
I joined Farcaster as a user in July 2022. Two year later, in July 2024, I joined Merkle to work on Farcaster.
Working on a social product I already loved alongside a strong team put together by @dwr and @v was a great experience. The difference in shipping velocity between Merkle and Google was night and day. Casting about what I shipped that week and the immediate feedback I received from our users was fun and motivating.
Funny enough, I decided to leave before I learned about the plans to sell. Perhaps I also sensed that it was a good time to close this chapter.
As for what's next for me, earlier this week I came back to work at Google Search to help build the native apps and AI Mode.
It was just a matter of time.
Google Now was launched 13 years ago (cards that showed your upcoming trip and such). It was the first incarnation of personal data being fed to users via Search.
Tech has come a long way since then.
https://x.com/i/status/2014374823787692485
I'm looking forward to Neymar's stewardship of Farcaster.
I've been a Farcaster user for almost four years, joining Merkle two years after I signed up. The community of builders and thoughtfulness is what kept me hooked.
I remember when @rish and @manan were gathering feedback for a creator centric Farcaster client before Neymar became what it is today.
This is a great outcome for both builders and users in the space.
This is a good list in terms of the tech and concepts specific to backend. I don't necessarily agree on going too deep on a stack but I'm biased since I've always been very stack agnostic. There is value on both depth and breadth.
However, a lot of this can be known and done by a Sr SWE.
At Staff level you are more equipped with dealing with ambiguity, evaluating trade-offs, prioritizing, designing systems with larger scope and/or impact, working with many stakeholders (customers/users, upstream/downstream dependencies, large cross-functional teams), etc.
In the age of AI these skills are significantly more important than knowing stacks deeply. You're still the ultimate decision maker though so be good as evaluating trade-offs and making decisions that are more future proof and that scale.
https://x.com/i/status/2013908799015498169
s/agent/swe
Coding with agents is not that different from having a team of SWEs.
Start with the design doc.
Don't context switch unnecessarily.
Have them review each other's code.
https://x.com/i/status/2013413295600902186
