James Mulholland

There is an ongoing debate about whether or not SaaS will be killed off by vibe coded alternatives. It's naïve to think 24hrs and a Cursor subscription is enough to replicate products that competent teams have spent years building. But it's also obviously true that tools like Claude Code are a crazy unlock. SaaS is hard. Vibe coding is real. Both can be true at the same time. What people are neglecting is that there's a middle ground. SaaS can be not just an app, but an extensible core. Vibe coded components can live on top of this core, without having to replace the whole thing. This means that the hard stuff like performance, scaling, and shared automation tooling—where experience and sweating the details matters—is handled by specialist teams. Then, vibe coding can take over in building the outer layers: integrations, custom features, and powering workflows that particularly matter to you personally. In the future, expect to see fewer, bigger SaaS products. MicroSaaS gets killed off by vibe coding. Big SaaS becomes more powerful as AI makes it easier to leverage an extensible core. The hard part is that SaaS is not extensible like this by default. If you wanted to modify [insert favourite SaaS app], you can't. Unless you're a developer at that company, you cannot modify the code. This needs to change. We need to build AI-friendly APIs and sandboxes so users really can write their own features. This is hard. Currently, Attio and Shopify are the only companies I've seen really execute on this vision. Getting this right is a crazy mix of security, DX, performance, interpreters and AI. But as I start to see people go from zero prior coding knowledge to fully working integration in <1hr, this vision is beginning to feel inevitable.

I've been getting lots of questions about how I use AI to work effectively on a production codebase. So, I'm putting together a course with everything I know. Agentic Engineering in Production aims to explain everything you need to leverage AI effectively in a real codebase. Less vibe coded greenfield todo app, more multi-million $ SaaS. Sign up for preview content here: course.productengineered.com

They didn't have to go so hard with the website

We’ve just done a big clear out at Attio. Lots of stuff that worked a year ago isn’t relevant. We’ve improved our understanding of what actually needs to be included. Another update is that we encourage engineers to generate their own skills on the fly, rather than necessarily putting everything in a shared context.

this is going to sound counterintuitive but perhaps companies should delete their cursor rules/agentsmd they have accumulated over the last year. treat markdown like cattle to be shot, instead of a pet that is named.

@chribjel It’s because politics and sports are bigger topics than the others. They scale proportionally to engagement

How do you even introduce this bug?

@LeRoyDesCimes how I mod my character's appearance in RPGs

Is Claude gaslighting me when it says it's not getting worse

@lukeknight The double Jude Law reduced the exclusive feel of the Legora ad for me, that's all!

@mulholo Made what feel desperate? I think the Legora ad was great Uber no one really talks about them

The real winner of all this AI stuff is Jude Law who’s in back to back Legora and Uber ads in Shoreditch, London today I reckon $5M plus for a few days work lol The man is cleaning up on tech companies urge to win The Legora ad was genius but I can’t say the same about Uber’s attempt!

@itunpredictable Wait until you hear about the butcher’s in east london that doubles as a listening bar

this is just boomer bait. this can't be real.

@mulholo at @attio on replacing the old, clunky CRM and giving every business software that feels like it was made for this era.

@weird_ceo @nicolasosharp @attio Working on some exciting things in this area! Are workflows the main one for you?

@mulholo @nicolasosharp @attio @mulholo

Joint #1 ;) We've put a huge amount of effort into making @attio the no. 1 CRM for agents, more to come!

@weird_ceo @nicolasosharp @attio Hi Constantine! We have lots of mutations available in the MCP today. Is there a particular one you're looking for?

@nicolasosharp @attio I love Attio, but MCP mutations are still very awaited :)

Two things are simultaneously true: 1. This incident was user error. Sandboxes, safe token usage and DB backups would have prevented this incident 2. System builders can take more responsibility. Well-designed tools can mitigate potential damage and help users move fast

No, not like that!

@the_mewc @attio Hey, just following up on this one! We struggled to replicate the issue here but do send more details if you run into errors again. We'd definitely like this to work!

@mulholo @attio Maybe it's just the oapi spec one

These days it's surely certainly in your interest to enable all AI bots to your docs? @attio would be nice

I get that this claim is in *some* sense true but it is more often the case that reliability and quality go hand in hand with speed, not at the cost of it

@Strakyo @attio We've found agents quite reliable at reading feature flag state from launch darkly and ensuring flags are enabled rather than toggled

@attio Nice workflow. The reliability hinge is idempotency between Attio events and LaunchDarkly writes; do you expose a retry-safe operation key per rollout?

Automate feature flag rollouts with our MCP. Pull customer requests from Attio, tell Claude to enable the right feature flags in LaunchDarkly, and roll out to the right customers automatically.

@DanielPrevoznik @attio Could you elaborate on what tagging records means here? List creation coming soon!

@mulholo @attio @mulholo - through some testing today noticed there is no way to tag records or create lists in the mcp. both would be useful! providing feedback in case helpful

i'd love if @attio mcp could be restructured to have less tools to call overall 🙏 34 is a lot

@DanielPrevoznik @attio Unfortunately, I don't think Cursor has yet implemented tool search. Would also be curious to hear how you're using the MCP in Cursor (custom app development?). We also provide a docs-specific MCP that you might be interested in docs.attio.com/mcp

thx for the response. this was helpful! part of my concern was related to cursor client imposing a cap on the number of active mcp tools (I thought it was like 80), so with attio and linear both enabled, I essentially couldn’t have another mcp server active with tools (since 34+41 tools). so overall less tools per server, means more active tools available across varied mcp servers to get work done. but as you mentioned maybe that’s no longer an issue. i have to check cursor again i can understand though with a crm how users would benefit from finetuned control over which tools have read / write access