Nilesh Bansal

Co-founder, ProdE AI - Software planning for agentic development. Like cursor, but for product managers.

[Nilesh Bansal @prode_ai]

@rauchg @vercel @v0 @nextjs Codebases are next. Query the code, and most planning meetings disappear.

[Nilesh Bansal @prode_ai]

@levie @levie For software companies, the institutional knowledge is the codebase, the highest fidelity source of truth. The ones who make it agent-queryable end-to-end win.

[Aaron Levie @levie]

This is effectively the #1 problem for AI agents in the enterprise.

As we go from agentic coding (where a large amount of context is in the code base, and users are technical enough to get the rest to the agent easily) to a world of knowledge work agents, the context problem becomes much more acute.

We see this every day with customers at Box. For existing digital knowledge, it’s often fragmented across legacy systems or environments that don’t play nice with agents, and have access controls that don’t map to the real work that needs to be done, which become a huge hurdle for getting agents the context they need. This has to all get moved to modern, secure cloud environments.

But also, companies often haven’t captured and digitized some of the critical context that agents need to work with. Decisions, processes, and workflows often live in people’s heads and tribal knowledge that need to get turned into unstructured data for agents.

This is actually one of the biggest points of leverage for applied AI companies, because they can work to specialize in getting agents exactly the information and domain expertise they need. But it’s also one of the reasons why FDEs and new system integrator plays will also work so well right now.

The companies that figure this out will be able to get the most out of AI going forward.

[Nilesh Bansal @prode_ai]

@levie @levie enterprise codebases run 100-200M tokens across hundreds of repos. No window fits. Even with Cursor and Claude Code pulling context automatically, they're working off a slice. The codebase IS the institutional knowledge. Most agents only see a fraction of it.

[Abhishek Bansal @bansalab93]

Claude Opus is getting more & more human!!

Giving superficial answers!
Asking users to confirm in code themselves!

And now, playing defensive and shifting the blame!

[Nilesh Bansal @prode_ai]

LLMs come and go, but your context is forever.

[Nilesh Bansal @prode_ai]

@jasonlk support bots not knowing their own product is a grounding problem, not an LLM one

[Nilesh Bansal @prode_ai]

@theo every dev tool eventually ships the feature that breaks your workflow

[Nilesh Bansal @prode_ai]

@theo model that overrides its own leash is not a feature

[Nilesh Bansal @prode_ai]

@rauchg @workflowsdk @pranaygp agents without retries are just demos with extra steps

[Nilesh Bansal @prode_ai]

@simonw benchmark a model on vibes long enough, the weights catch up

[Nilesh Bansal @prode_ai]

@GergelyOrosz the real model update is the toggle they moved

[Nilesh Bansal @prode_ai]

@GergelyOrosz models don't regress. defaults do. silently.

[Nilesh Bansal @prode_ai]

@emollick all the memory hacks are just external context graphs in a trenchcoat

[Nilesh Bansal @prode_ai]

@Polymarket human in the loop is just the loop now

[Nilesh Bansal @prode_ai]

@emollick denominating compute in coffee is the kind of unit my PM brain needed

[Nilesh Bansal @prode_ai]

@emollick the cry wolf cycle makes it impossible to notice when it actually works

[Nilesh Bansal @prode_ai]

@dani_avila7 @bcherny 7.7 billion tokens and still shipping. the bottleneck was never the code

[Nilesh Bansal @prode_ai]

@kloss_xyz context management is the unsexy infrastructure nobody wants to build but everyone needs

[Nilesh Bansal @prode_ai]

@jasonlk @buccocapital shipping late because nobody spec'd what the AI was supposed to replace