VCs Fund Context Layer Tools to Enhance AI Coding Assistants

I don’t know if there will be any big winners here, or coding harnesses will just make it dead simple to add in context or discover them (MCP is a two in this direction btw).

Also kind of reminds me about the “internal knowledge base solution” trend ~2015-2022 also

@GergelyOrosz you mean loads of people building lore?

https://github.com/itsthelore/rac-core

@GergelyOrosz Same for "if only Claude Code was reliable" 😅 Trying to fix a fundamentally unreliable thing

@GergelyOrosz there’s a big market here.

I’m not convinced coding harnesses will get all the way there though. engineering teams already spread work and workflows across wikis, Jira, Asana, and tons of other tools. There’s a lot of systems to hold in mind and MCP would bloat pretty quickly

@GergelyOrosz "Maybe anthropic will buy us" is rather risky strategy.

@GergelyOrosz Knowledge based systems, uni, year 2003 :) problem: how to keep it up to date, humans are lazy. Everything goes via AI and every step/decision is saved into memory? But then you are quickly in the “AI must hear everything” territory

@GergelyOrosz My experience is that institutional knowledge in large corps is very fragmented and unstructured. It may be possible to build such a layer of knowledge and give Claude the context, but the getting the correct training data could be a real challenge.

@GergelyOrosz Do you also see the demand from normal companies who would be their customers though?

I don't...yet

I believe it's a real problem but most businesses are at least a year behind it actually being real problem for them

@GergelyOrosz Startup I was working at last year built that as a generic solution, super capable and easy to use.. but failed to raise past the initial angel phase.

Pre claudcode unfortunately so we built the actual engine first and money ran out before the UX was in place.

@GergelyOrosz seems like the kind of thing that will become business as usual work for all eng teams to maintain, since its similar to a companies data pipelines, custom dev needed

@GergelyOrosz I'm probably the only one who did not build a ai solution for this, but actually focused on putting the right context at the right time in front of the engineers, secure, open source & deterministic.

think dynamic bookmarks

and which can be used by claude code

tooljump dot dev

@GergelyOrosz everyone's building the layer that reads all your systems. the catch is the same as the pitch: it only works once a team wires every system into one startup's box. value and adoption problem, same fact.

@GergelyOrosz Back to ESB (Enterprise Service Bus) Maybe ASB?

It doesn't suffice, loosely connected systems are far from sufficient, at best for narrow use cases even with agents. You need to also lap out what people really do, all variants of fragmented processes and that is then a realistic starting point for agents in business. Everything else is doomed to fail except for narrow cases.

@GergelyOrosz > MCP is a two in this direction btw Not sure I understand. 2 as in 2 out of 10 as in "hey it's a step but it's not great". or ... something else?

@GergelyOrosz I remember when Glean launched (2021?) and I thought this would be the next biggest Saas.

@GergelyOrosz Think the hot topic being the enterprise context layer recently and the step beyond the system into company culture/op model etc... could be driving that, you need that system knowledge first no? In theory. That and shitty data catalogues in big firms

@GergelyOrosz I saw one that built a Stack and a Brain

@GergelyOrosz THIS is the hot new thing.

I imagine the winners will be whoever currently has distribution with their offerings and has a seamless way to value add with a "context layer" with very little additional work

eg

https://aws.amazon.com/blogs/machine-learning/context-intelligence-for-your-data-and-ai-agents-at-scale/

@GergelyOrosz technical writers (or whatever title we give them by then) comeback scheduled for 2027