Craft Docs resisted AI hype as a nimble, well-run startup – but has just made a sharp pivot to AI, building a universal agentic tool everyone there now uses. The results are phenomenal. Exclusive.
0/ This is reminiscent of all these people building personal life management systems in Obsidian :)
1/ Agent orchestration and communication with humans will be the theme for 2026. Finding the right level of integration: yesterday's manual point and click, to something like Craft, all the way to Gas Town.
2/ This past week I saw how hard it was to convince even professional SWEs with real experience how valuable agentic coding is. Seems like no amount of articles or other people's experience will help.
Interesting, although I wonder how safe is it to generalize Craft's case to other larger companies, where the management is less engineering oriented and more business centered, and focused on seeing only the cost savings aspects of AI coding tools.
What do you think, is this a lone success story, or do you see other trends in other companies ?
Moutaz: great question! Feels to me this can only happen with either founder-driven startups (the conviction came from the founder, plus the mandate) or places where the CEO is so hands-on and can tell when AI *really* works, having built key software for the company.
In this case, Balint wrote most of v1 of Craft, and most of Craft Agents. The question is: how typical or rare is this? I think very rare… but it should be more common!
Thanks for sharing the story and for your reply. I am a software engineer working in Germany and I must admit that I am trying to cope with the hype but also trying to adapt and learn to professionally position myself correctly for the future. I will try to _market_ the insights and lessons from this story in my workplace and see if we can get somewhere. Thanks again ...
Chi Ho: asked Balint, and they are most likely going to land on Plain (https://www.plain.com/). They looked at a ton of alternatives and like Plain the most so far. (as usual, I have no association, and the move has not yet happened)
The pace of change in this industry is crazy. I have been working on some less code intensive and more design intensive projects since ~November (then MCP server, now IaC migrations) and I'm speechless about what's happening.
I'm feeling confused about it since even though I'm in a tech lead/architect role, I love coding and basically within a few weeks, I need to change my and my team's views about how work should be done, despite how we've been doing things for years or even decades (and enjoying "the old ways"!).
The only thing that caught my attention is that we're talking about engineers and engineering processes, but categorize people as either "believers" or "non-believers" like this was some sort of religion. Odd!
Maybe we'll come up with better terms over time and also be able to observe real productivity measurements instead of anecdotal "10x gains" which also rings a bell for me (think "10x developers").
I have filled out the AI survey, eager to learn about the results!
I wonder if Balint would have started on Agents if he knew Claude Cowork was coming out - it feels like there is a lot of overlap in the direction and capabilities.
(Amazing article, this is the good kind of FOMO, cheers on a super interesting tool!)
One advantage that Craft Agents seems to have right now is that, because it is built on Electron, it can run on a windows machine, while Claude Cowork is currently mac-only. I'm sure that will change for Cowork within the next couple of months, but still...
So much to relate to here:
0/ This is reminiscent of all these people building personal life management systems in Obsidian :)
1/ Agent orchestration and communication with humans will be the theme for 2026. Finding the right level of integration: yesterday's manual point and click, to something like Craft, all the way to Gas Town.
2/ This past week I saw how hard it was to convince even professional SWEs with real experience how valuable agentic coding is. Seems like no amount of articles or other people's experience will help.
Very good overview of a thoughtful AI-implementation. I hope we trend further towards this avoidance of "Copilot everywhere" haha
Interesting, although I wonder how safe is it to generalize Craft's case to other larger companies, where the management is less engineering oriented and more business centered, and focused on seeing only the cost savings aspects of AI coding tools.
What do you think, is this a lone success story, or do you see other trends in other companies ?
Moutaz: great question! Feels to me this can only happen with either founder-driven startups (the conviction came from the founder, plus the mandate) or places where the CEO is so hands-on and can tell when AI *really* works, having built key software for the company.
In this case, Balint wrote most of v1 of Craft, and most of Craft Agents. The question is: how typical or rare is this? I think very rare… but it should be more common!
Thanks for sharing the story and for your reply. I am a software engineer working in Germany and I must admit that I am trying to cope with the hype but also trying to adapt and learn to professionally position myself correctly for the future. I will try to _market_ the insights and lessons from this story in my workplace and see if we can get somewhere. Thanks again ...
Any idea (hints?) what ticket system they are evaluating and/or moving to?
We are in the same boat, we use both Zendesk and Freescout and they aren't API-first, where we can tack on a "proper" AI agent.
Chi Ho: asked Balint, and they are most likely going to land on Plain (https://www.plain.com/). They looked at a ton of alternatives and like Plain the most so far. (as usual, I have no association, and the move has not yet happened)
Nice writeup!
The pace of change in this industry is crazy. I have been working on some less code intensive and more design intensive projects since ~November (then MCP server, now IaC migrations) and I'm speechless about what's happening.
I'm feeling confused about it since even though I'm in a tech lead/architect role, I love coding and basically within a few weeks, I need to change my and my team's views about how work should be done, despite how we've been doing things for years or even decades (and enjoying "the old ways"!).
The only thing that caught my attention is that we're talking about engineers and engineering processes, but categorize people as either "believers" or "non-believers" like this was some sort of religion. Odd!
Maybe we'll come up with better terms over time and also be able to observe real productivity measurements instead of anecdotal "10x gains" which also rings a bell for me (think "10x developers").
I have filled out the AI survey, eager to learn about the results!
Thank you for sharing this interesting journey.
If I understood correctly, he built an app on top of the Claude Code engine that is able to "talk" to external data sources and MCPs.
I have a question about this:
- Why create this app when it could be done via connecting the different existing MCPs to Claude Code directly?
- What value brings that "wrapper" (I don't want to be disrespectful here)?
Thanks!
One more great article 👏 I hated the matrix variant of the craft agents however 😆
I wonder if Balint would have started on Agents if he knew Claude Cowork was coming out - it feels like there is a lot of overlap in the direction and capabilities.
(Amazing article, this is the good kind of FOMO, cheers on a super interesting tool!)
One advantage that Craft Agents seems to have right now is that, because it is built on Electron, it can run on a windows machine, while Claude Cowork is currently mac-only. I'm sure that will change for Cowork within the next couple of months, but still...