Hacker News

Article URL: https://software.rajivprab.com/2019/12/19/when-feature-flags-do-and-dont-make-sense/

Comments URL: https://news.ycombinator.com/item?id=49160383

Points: 1

# Comments: 0

Hello HN,

I’m Dallin, building Keystroke with my co-founder Blake. Keystroke is a recent pivot from Buster, our original YC W24 company.

Keystroke lets you build agents and durable workflows as TypeScript in your own repo, then deploy them to a web application where your team can use them. Our repo (ELv2 licensed) is here: https://github.com/keystrokehq/keystroke.

At Buster, we spent a lot of time building internal agents and automations. We used visual platforms like n8n and Zapier, as well as code-first tools like Trigger.dev, Mastra, and several of Vercel’s open frameworks. We genuinely liked many of them, but kept encountering two different problems:

1. Visual platforms gave us integrations, credentials, hosting, and an app the rest of the team could use, but became difficult to test, review, and debug as our systems grew. 2. Code-first tools gave us the flexibility of ordinary TypeScript and worked well with coding agents, but left us building the UI, runtime, permissions, credentials, and operational tooling ourselves.

So, we started building the thing we wish existed: a code-first framework that made it easy to build complex agents and AI systems, with a web app for teams to collaborate on top. We liked it so much that we decided to pivot away from Buster and focus on building Keystroke full time.

With Keystroke, I was able to rebuild our W24 idea (an AI data analyst) in ~5 minutes. I pointed Cursor at our old Buster repo and it was able to:

- spin up a local Keystroke project - rebuild the OG Buster agent w/ memory, its own file system, etc - connect our Postgres DB, PostHog, Metabase, and Slack - build skills for using each of the data sources - explore our DB and add schema docs to the file system - build a workflow to keep schema docs synced - name the agent "Delbert" - then pushed him up to the Keystroke app

Since then, “Delbert” has been answering ad-hoc questions for me in Slack on a regular basis. He builds me metabase dashboards when I need them. He sets his own triggers, reviews key metrics periodically, and pings me in Slack if anything looks off.

In Keystroke, workflows are ordinary async TypeScript: a `for` loop is an actual `for` loop, inputs use Zod schemas, and everything you build can be tested with Vitest. Our managed runtime makes workflows durable across errors, sleeps, approval steps, and process restarts.

Agents wrap Vercel’s AI SDK and can use workflows, actions, other agents, MCP servers, and integrations as tools. They also include memory, a persistent filesystem, web search, triggers, and optional VM sandboxes.

Deployed agents and workflows instantly appear in the web app with a chat UI, workflow visualizations, input forms, sharing, credential management, integrations, and logs.

The entire platform was built from the ground up with coding agents in mind. Your coding agent can use the CLI to scaffold a local directory with a Keystroke project and an *AGENTS.md*. Then, it can build whatever agent or AI system you want, search our docs, run tests, and push what you build up to the Keystroke platform.

We plan to make money from our cloud offering, which has a free plan and usage-based pricing.

Our docs are here: https://keystroke.ai/docs.

You can watch a quick demo of Keystroke here: https://supercut.ai/share/keystroke/zeor1pD0bNlSTOuLUWmRBT?v....

You can try on cloud for free at: https://keystroke.ai/.

We’d especially appreciate feedback from people who have maintained agents and AI systems in visual tools or code-first frameworks. Where does this approach seem useful, and where are we rebuilding something that already works well elsewhere?

Comments URL: https://news.ycombinator.com/item?id=49160376

Points: 1

# Comments: 0

Article URL: https://tucson.com/news/local/crime-courts/article_9ae8e2d7-1db8-4c86-9dee-7c439f4ea7c0.html

Comments URL: https://news.ycombinator.com/item?id=49160359

Points: 2

# Comments: 1

Article URL: https://twitter.com/MaseehG_/status/2084327646209392907

Comments URL: https://news.ycombinator.com/item?id=49160354

Points: 3

# Comments: 0

For the last 2 weeks I have been running Codex + GPT 5.6 Sol Ultra non-stop for almost 13 days, working on a huge extension to my SaaS product / business.

I have done AI coding of huge features in the range of 50-60k LOC before, which were massive already, we managed to ship them and actually deliver value.

I know this sounds borderline ridiculous, but this time I don't know what to do.

What are the next steps here? It is obviously impossible to review, but it is almost as impossible to actually test. I had put A LOT of work into planning and the workflows and features were spec'ed very precisely and with lots of care and analysis. So far I have only found cosmetic issues, the entire core of the application seems to be working just fine.

I have thought about starting with a user manual / knowledge base for it, to capture the features and have plenty of documentation. But then what? Launch it to production?

Have I passed the threshold of what is humanly possible to do with AI coding, or is there any hope?

Comments URL: https://news.ycombinator.com/item?id=49160339

Points: 1

# Comments: 0

Article URL: https://medium.com/@kisnorbert87/gravity-is-not-a-force-its-an-error-correcting-code-68a48374e575

Comments URL: https://news.ycombinator.com/item?id=49160335

Points: 2

# Comments: 0

Article URL: https://willdepue.net/writings/a-stargate-for-data/

Comments URL: https://news.ycombinator.com/item?id=49160332

Points: 1

# Comments: 0

Article URL: https://phauer.com/2018/moving-back-from-gradle-to-maven/

Comments URL: https://news.ycombinator.com/item?id=49160286

Points: 1

# Comments: 0

Article URL: https://github.com/ConorsCode/open-jobs-data

Comments URL: https://news.ycombinator.com/item?id=49160283

Points: 1

# Comments: 0

Article URL: https://gcc.gnu.org/ai-policy.html

Comments URL: https://news.ycombinator.com/item?id=49160279

Points: 1

# Comments: 0

Common characteristics of these articles:

- Promote fake AI. Talking advantages but no or very few disadvantages

- No clear central argument, no clear and right evidences, no rigorous logic, no values for readers

- Generated by fake AI or LLM

Comments URL: https://news.ycombinator.com/item?id=49160266

Points: 2

# Comments: 0

Article URL: https://zerotrustintelligence.io/verified-done.html

Comments URL: https://news.ycombinator.com/item?id=49160254

Points: 1

# Comments: 1

Hey HN - Mert here, maker of MoneyCo (thecompany.money). MoneyCo connects to multiple Quickbooks and Xero accounts by default and lets you do your bookkeeping using a coding agent.

I got here after talking to multiple accountants and hearing them struggle to use Claude, Codex, Hermes and OpenClaw. The difficulties they had were classic:

1. QBO and Xero MCPs are brittle and allow limited access 2. They were only able to figure out how to run these locally 3. LLMs not being good at math 4. Configuring client memory

Coding agents disguised as one that does accounting felt like a simple yet powerful idea as a solution:

1. We connect to QBO and Xero using API instead of MCP and make it easy for multiple accounts connection 2. We deploy all client files along with `pi --rpc` in a sandbox so all can work on the cloud 3. A coding agent can simply write code for deterministic computations and run tests to catch inaccuracies 4. Each chat made progressively gets stored in the persistent sandbox meaning that agent can search through older conversations to look into its memory

I had to rebuild some of the architecture here several times as it required building a durable, persistent cloud coding agent essentially. Each client is configured almost as a repo/filesystem of its own that lives there forever, including the agent chats made by the users.

Stack is

- django for backend - modal for sandboxes - streaming uses two paths (1) for livestreaming (2) for durable background ones. Both use Cloudflare durable objects - frontend in vite - agent is a fork of pi - no MCP usage at all. For OCR, memory search, IRS docs search and further things, there is an embedded cli called `tlmc` that agent is encouraged to use in its system prompt

HN is not the right place to launch an accounting product however I thought the way this is built might be relevant to cloud agents and how to architect them.

Please let me know any feedback, requests and most of all any brutal feedback!

Comments URL: https://news.ycombinator.com/item?id=49160189

Points: 1

# Comments: 0

Article URL: https://github.com/arvidma/dataclasses

Comments URL: https://news.ycombinator.com/item?id=49160160

Points: 1

# Comments: 1

Article URL: https://www.thehistoryblog.com/archives/15375

Comments URL: https://news.ycombinator.com/item?id=49160125

Points: 1

# Comments: 0

Article URL: https://notawebsite.fun

Comments URL: https://news.ycombinator.com/item?id=49160121

Points: 3

# Comments: 0

Article URL: https://www.rnz.co.nz/news/business/874814/many-businesses-using-ai-to-save-time-but-do-they-know-what-they-re-doing

Comments URL: https://news.ycombinator.com/item?id=49160101

Points: 4

# Comments: 0

Article URL: https://medium.com/@alanscottencinas/business-impersonation-fraud-when-trust-is-the-target-1f48826b647b

Comments URL: https://news.ycombinator.com/item?id=49160092

Points: 1

# Comments: 0

Article URL: https://arxiv.org/abs/2607.29005

Comments URL: https://news.ycombinator.com/item?id=49160091

Points: 1

# Comments: 0

Article URL: https://thezvi.substack.com/p/openais-unreleased-model-astra-solves

Comments URL: https://news.ycombinator.com/item?id=49160081

Points: 2

# Comments: 0