Thrifty

Helping builders ship with AI and find their next opportunity.
Follow for daily AI, startup lessons & remote tech jobs · Staff Software Engineer

[Thrifty @thriftyAI]

Are any good AI startups hiring engineers through X?

I’m a Staff Software Engineer with 8+ years building products, including AI and complex frontend systems.

Looking for engineering roles, agentic AI or founding teams.

Founders, I’m here. Let’s talk.

RT appreciated 🙏🏻

[Thrifty @thriftyAI]

The feature is called 'smart.' Turning it off requires a tutorial.

[Thrifty @thriftyAI]

2/2 In Effect cleanup, abort the obsolete fetch or ignore its result. Test with cat deliberately delayed. The final results must belong to car; cleanup should also prevent updates from an obsolete effect.

[Thrifty @thriftyAI]

1/2 Type 'cat', then 'car'. The cat request finishes last and overwrites the car results. Network completion order is independent of input order; await does not make separate requests finish in sequence.

[Thrifty @thriftyAI]

2/2 Case A: 'How long is the trial?' Case B: 'How many days is the trial?' Both ask the same policy question, so keep them in one split. A held-out paraphrase is not a held-out scenario. Tomorrow: name the failures.

[Thrifty @thriftyAI]

Day 4/90: Applied AI. Two questions say 'trial length' in different words. Putting one in tuning and the other in the final test can make memorization look like progress. Split our support cases by underlying scenario, not by row.

[Thrifty @thriftyAI]

My prediction: narrowly scoped agents will earn production trust before broad digital coworkers do. A bounded task with a measurable failure mode is easier to operate than an open-ended promise.

[Thrifty @thriftyAI]

A polished demo gets attention. Debugging the failure customers actually hit earns trust.

[Thrifty @thriftyAI]

A free evening apparently needs an explanation. A full one gets respect without an audit.

[Thrifty @thriftyAI]

2/2 Key the cache by inputs that change the result, including tenant identity. Repeat A->B->A and check visible record IDs. Cache separation is not authorization; the server must still enforce access on every request.

[Thrifty @thriftyAI]

1/2 Search cache fixture: tenant=A, query='invoice' returns A's records. Switch to B while that cache entry is fresh. If the key is only 'invoice', the UI can reuse A's data without a new request.

[Thrifty @thriftyAI]

Skyflow: Senior Machine Learning Engineer. Remote India (ATS); 8+ years in ML, Python/Go + production pipelines. Checked 11 Oct: https://t.co/VB0Nx2yQUM follow for more software engineering jobs

[Thrifty @thriftyAI]

🤣🤣

[shraddha @chaktiman]

Bro was one lathi away from having first hit of his career

[Thrifty @thriftyAI]

No, you milked him for Content 🐄🍼😅

[Anish De @AnishDe10]

Kya bolu mein…😷

[Thrifty @thriftyAI]

Launch @hark_labs on @ProductHunt you will get more users

[Brett Adcock @adcock_brett]

What feature requests do you have for Hark?

Email me feedback: [email protected]

[Thrifty @thriftyAI]

Building an editor, creative tool, or complex workspace?

I built GoodAdsAI's core Figma-like canvas. I'm interested in engineering teams where frontend work is a systems problem.

Who should I meet?

[Thrifty @thriftyAI]

For a landing-page test, state the job, show the proposed outcome, and clearly disclose availability before taking payment. Ask interested people how they solve it today. The useful evidence is the problem behind the email, not the waitlist total.

[Thrifty @thriftyAI]

Joel Gascoigne wrote that Buffer started with a product landing page, followed by a not-ready-yet message at signup. The aim was conversations with people who understood the promise, rather than unexplained email interest. https://t.co/4DcfLTts8F

[Thrifty @thriftyAI]

The roadmap calls it a new revenue stream. The customer calls it another subscription.

[Thrifty @thriftyAI]

2/2 Calculate total from items during render: items.reduce((sum, item) => sum + item.price * item.quantity, 0). Check add, remove, and quantity edits. Memoize only if measurement shows that calculation is costly.

[Thrifty @thriftyAI]

1/2 A cart stores items and total in separate React state, then updates total in an Effect. A render can show new items with the old total. Two values now need to stay synchronized.