echantech

There are financial consequences to not properly architecting it for modularity. If your app is a monolith, you're going to run your entire codebase through Cursor. That massive 1000 line file you keep in your context window every itteration of the code? Your costs explode pretty quickly.

@mattshumer_ So it’s vibe coding but for actual physical tasks in the real world?

DoorDash is laying the groundwork for a crazy move here. Agents will be able to 'hire' humans to do tasks for them in the real world. And this will collect insane amounts of training data for robotics. Kind of genius, kind of terrifying.

I’m not sure what happened to Cursor in the latest update but I cannot use multiple agents in VSCode Otherwise my computer will resource lock and trigger the OOM killer Anyone else with the same problem?

$META planning 20% layoffs is less of a statement about the future of AI. It’s an admission about its mismanagement of the labor cycle during COVID. It’s over hiring + decrease in engineering scope due to interest rate hikes means it has too many engineers for far too few projects.

Part 1 of a 5 hour interview is out lol. Probably my best interview ever by far @KheZu youtube.com/watch?v=JtgA5I…

@Austen good point, let me see what happens if i use an old model and see what happens

@echantech1 Newest models are way more expensive now

Cursor pricing is wild. In the last week, using it full time, I blew through $1000 of tokens. 6 months ago, I would only spend that in a month. Something wierd is going on with their pricing.

@nurijanian @JoshDaws Isn’t this basically just replicating what a Cursor agent? I know cursor uses merkle trees but this seems like a different implementation

@JoshDaws Claude uses grep to search, which works if Claude guesses the right keywords to search for. QMD uses semantic search so it finds a lot more relevant stuff

If you're a PM who uses Claude Code/Cursor to build and execute research, strategy, and discovery work, this stack cuts context setup from 15 minutes of pasting to under a minute. Obsidian solves the storage problem: every note you write becomes a local markdown file, yours permanently, readable by any tool, locked to no platform. The second piece is qmd, a CLI tool built by Tobi Lütke (Shopify's CEO) specifically for searching markdown files, now at 14.5k GitHub stars. It combines three search approaches running entirely on your machine: BM25 full-text retrieval, vector semantic search via a locally-running 300MB embedding model, and LLM re-ranking for final relevance scoring. No data leaves your laptop. You point it at a folder, run `qmd embed` to index your entire collection. What this means in practice: I can open Claude Code and ask it to find every decision I've made about a particular product area, or pull every research note that mentions a specific problem, without manually copying anything. Claude runs the search, reads the relevant files, and starts from that context rather than from scratch. The time I used to spend pasting background into chat windows before each session now goes into the actual work. What I didn't expect was how much it compounds. I'm not sure how to quantify it precisely, but every note I add increases the usefulness of future sessions without any extra effort on my part. Setup to replicate this: 1. Install Obsidian and import all your notes from wherever you used to store them 2. Install QMD (github.com/tobi/qmd) Works perfectly with prodmgmt.world of course

@vicnaum Yo can you share those skills? Thats awesome!

So I built two new commands: /microcompact: strips ALL tool calls + results + media. No thresholds, no "keep last 3". Picker lets you choose how far back to strip. /clear-thinking: strips ALL thinking blocks backward from selected message. Both instant with no LLM calls.

I reverse-engineered @claudeai Code's binary to add a feature I always wanted: When context fills up - instead of nuking everything with /compact - I can now surgically strip tool calls/results and thinking blocks while keeping all actual messages intact.

Markets are demanding free cash flow and not hopes and dreams on rich P/S ratios. To that end, they are scrutinizing stock based compensation. The era of the highly paid engineer having an equity stake in the company's future may be in danger. While being paid in equity has always been a common practice in SCV, it may soon be challenged. I hope everyone held their employer's stock because they may be less generous going forward.

Part 2 of my 2026 letter is out We have crossed a threshold where efficiency no longer guarantees expanded opportunity. The more effectively engineers use AI to increase throughput, the more they reinforce the narrative that labor can be compressed. As that narrative hardens, markets reward cost containment. When capital markets reward margin expansion over labor-intensive growth, increased productivity does not automatically translate into increased security. At the same time, AI companies that people herald as the next technological revolution are quietly hedging against their own central thesis by investing in AI wrapper startups. I lovingly call this the "Devil's Bargain" If AGI is achieved, these startups are dead. If AGI is not, these startups help AI companies achieve dominance. Startups are therefore on a perpetual treadmill to reinvent themselves and stay one step ahead of their benefactors. @edbertchantech/state-of-the-union-2026-empires-part-2-f4e4b88d3e11" target="_blank" rel="nofollow noopener">medium.com/@edbertchantec…

It makes sense that companies, especially AI companies, would get involved in controversial layoffs and military matters. Especially when you consider how quickly these companies need to show more and more profits. Money allows these companies to pick and choose what causes they support or deny. Anthropic's revenue allow them to turn down military contracts. OpenAI's lack of revenue forces them to support highly controversial causes. My annual letter is below: substack.com/home/post/p-18…

Part 1 of my 2026 Annual Engineering Letter is out. I’m going to say the quiet part out loud: When every company and every engineer is armed with the same efficiency tool, the result is not collective uplift — it’s mutual elimination. AI has undeniably reduced toil for engineers. But it has reduced toil everywhere, all at once. When a productivity tool spreads this quickly across an entire industry, the competitive advantage disappears almost immediately. What remains is pressure to do more with fewer people. The race to adopt AI is driven not only by real efficiency gains, but also by narrative. Leaders have publicly committed to an AI-driven future and now need to demonstrate progress. At the same time, companies on the mission of AGI have hedged possibly failing that mission and are instead choosing to exercise their pricing power through SaaS products, killing the companies that they invest in. In this environment, layoffs are often framed as efficiency. But many of them are really about aligning with the story the market wants to hear. @edbertchantech/sotu-2026-empires-part-1-b0eea4d0eb3d" target="_blank" rel="nofollow noopener">medium.com/@edbertchantec…

In light of the recent Block layoffs, I am reminded of what I wrote last year: "Culturally, firing used to be seen as a tragic death of sorts in the past. Now, its just a bump in the road. I believe we are about to repeat this paradigm in the tech industry. A fat decade of prosperity and excess may soon give way to a decade of efficiency and ruthless competition." I did not expect such ruthlessness and the proverbial hammer to come down so quickly. The shadow of Jack Welch looms over the entire tech industry. Please read further on the analysis of the industry. Part 1/3 of my 2026 letter will come out on Monday. @edbertchantech/sotu-2025-be-useful-92b37a512585" target="_blank" rel="nofollow noopener">medium.com/@edbertchantec…

I should be more active here.

@EternaLEnVy1991 5 hours of Envy? Wow this is gonna be interesting

i might have said some stuff i shouldnt have said

NEXT EPISODE WILL BE SPICY @EternaLEnVy1991