Ray Ploski

We built CoreStory because legacy code is the #1 blocker for enterprise AI — and nobody was solving it right. I talked about it on the Innovators Podcast: youtu.be/XFf1DzxqZBU?ut…

AI coding agents fail on complex tasks for one reason: poor context. Bigger context windows don’t fix it. They dilute attention and hide root causes. Stop feeding raw code. Start engineering context. --> corestory.ai

I watched a guy run a sub-six-minute mile pace ... uphill ... on skis while I ate a Reese's Blizzard from Dairy Queen (with extra Reese's)

9. Use Claude for data & analytics Ask Claude Code to use the "bq" CLI to pull and analyze metrics on the fly. We have a BigQuery skill checked into the codebase, and everyone on the team uses it for anlytics queries directly in Claude Code. Personally, I haven't written a line of SQL in 6+ months. This works for any database that has a CLI, MCP, or API.

We’re excited to announce that CoreStory has partnered with @FactoryAI. We’re enabling AI agents to work with real architectural context, driving more accurate, reliable, and production-ready code across the SDLC. Read more: corestory.ai/post/corestory…

Enterprise SWE agents fail without code intelligence. CoreStory → 51% accuracy boost for SWE agents = The difference between "doesn't work" and "ships to prod" Legacy analysis: 18 months → minutes Result: Enterprise SWE agent dev that actually works corestory.ai/post/specifica…

New from @CoreStoryAI: Engineering Monthly — a regular look into the ideas shaping our engineering community. • Spec-driven dev in the AI era • SWE-Bench + real-world • Modular, skill-based agents Read → corestory.ai/post/corestory… #AI #CodeIntelligence #SpecDrivenDev

@NEA @corestoryai Legacy software is one of the biggest untapped frontiers for AI. @corestoryai approach, turning decades of code into living, structured specs, could unlock trillions in enterprise value trapped in technical debt.

That’s a wrap! 💥 At #GitHubUniverse 2025 our founder showed how @corestoryai makes Copilot smarter by design — specs as context, full automation, & 3× ROI for enterprise devs. Slides 👉hubs.la/Q03QRSDW0 #SpecDrivenDevelopment #AI #CodeIntelligence #CoreStory

Thrilled to be part of this journey. With $32 million in new funding, we’re scaling a platform that cuts modernization effort in half and gives enterprises living specifications they can trust for decades to come.

Portland on brand as ever oregonlive.com/portland/2025/…

i can't believe this shit actually works

Spec-Driven-Development is a reality. Pull your spec out of existing source and push it into the coding agent of your choice. It's the next chapter in code modernization.

At CoreStory, we help ensure those agents don’t drift from business intent.💡 Specs generated + kept alive from natural language requirements 🏢 Enterprise alignment at scale GPT-5 Codex + CoreStory = the next era of Spec-Driven Development. #CoreStory #AI #SpecDrivenDevelopment

🚀 CoreStory is live! We’re bringing AI-powered code intelligence to the enterprise—modernizing legacy systems at scale. Here’s what the press is saying 👇

Introducing CoreStory: the platform that translates complex codebases into natural language requirements—accelerating modernization at scale. The future of requirements-driven development starts here. 🌐 corestory.ai #CoreStory #AI #RequireIntelligence

@ryrobes The company @corestoryai literally builds out spec for existing systems - whether vibe coded or legacy for this purpose. GitHub Co-Pilot accuracy increases over 50% when using it in this type of use case.

A lot has been said of sustainability of "vibe" coded apps (and it's not wrong). But the truth is, if you have really good functional requirements docs & good automated feedback loops - a lot of these issues can be dealt with. Entire modules can be rebuilt as needed - and if you are proficient with the language it is writing (important! - otherwise you might be that guy with a tattoo in Chinese on your arm that you were told says "Badass", but it really means "Horrible Flatulence") - you can shepherd and insert implementation hints in the FRD. This deals with many of the complexity snowball and sliding "comprehension window" issues. This is also made infinitely easier with a functional-first lang like Clojure where entire sections can be pure. Yes, even with non-trivial applications. Keep modules as isolated as possible so they can be worked on in a vacuum - and make the main runstream understandable without having to understand all the modules (again, pure functions and straightforward side-effects help a lot). You can get a lot done in a 300k token context window if it is well defined, isolated, and given clear direction. Get good at writing really detailed technical explanations - and get comfortable with lots of iterations - and feed the learnings (failures & successes alike) back into the FRD. Maybe version 4 is a complete shit-show - but it WORKS end-to-end and proves viability. Great, than becomes an archaeological artifact to refine the new FRD for version 5. You become the PM/EM/QA/Designer (screenshot mock ups in something like TLDraw are super helpful if you don't want startup bootstrap slop). Bottom line is that if you know what you're doing (i.e. COULD HAVE written it on your own) - your chances of success go way up. A bitter pill perhaps - but with the state of the models, that's where we are at.

LLMs are prediction machines—great at patterns but lack true intelligence. Enter #IntelligenceModels by @Crowdbotics: structured reasoning + business logic = reliable AI-driven outcomes. #AI #MachineLearning #TechInnovation

< 10% of How I AI viewers on youtube are female 🆘 When @yourgirlhils and I talk about women getting left behind in the AI boom, this is what we mean. My darling ladies in tech and not-tech, what can I do to make AI more accessible, relevant, fun, and applicable to you?

@GregSharples @GergelyOrosz Our ability to understand codebases is being rapidly outpaced by our ability to generate new code. It is a problem that we are solving for brownfield and new, greenfield vibe coded apps today at @Crowdbotics. We would love to hear your ideas on how best to serve the industry.

@GergelyOrosz Agreed, and I think the dev version of vibe-coding is basically just what coding is now for devs (or will be soon). If you know what’s happening and are reviewing etc I don’t think there’s any compelling reason not to

A thing I notice: “vibe coding” tends to mean something substantially different with devs vs non-devs saying it. Non-dev saying they “vibe code:” they use prompts only. Dev saying they “vibe code:” oftentimes it’s “the AI wrote *most* of my code.” Big difference!