Himanshu

5.6K posts

Himanshu banner
Himanshu

Himanshu

@nothiingf4

Building agents | Interested in shipping fast opportunities | Medium: https://t.co/x2bjY0DnEj

Katılım Mayıs 2024
783 Takip Edilen2.5K Takipçiler
Avid
Avid@Av1dlive·
Stop Swiping on Tinder Today I built Tinder for startup founders and it took me less than an hour Swipe on founders in your niches. Match. Get icebreakers. DM without cringe Apps like SuperappAI make building ios apps super easy
English
6
4
9
275
Himanshu
Himanshu@nothiingf4·
morning fellas ;) - pulled an all nighter again - made a few more pr's still i need to test - learnt a new thing about the memory thing in teh agentic space - been building a triage system and trying to add a few cool features in it Don’t think of cost, think of value ~ nothiingf4(something which i follow)
Himanshu tweet media
English
4
1
32
450
Siddharth
Siddharth@Pseudo_Sid26·
Going super busy with the new work. Ain't getting any time. From the past 2 days, am like working 14-15 hrs a day. -Raised a PR on second day and merged. Yes like every other place, the first thing I do is implement langfuse. It was already there but it was so shit, fixed the trace and nest pattern inside langfuse and shipped the PR. - Reviewed and made some initial small architectural suggestions for another sub project. - Made an eval/beta benchmark test setup for a new model release for an org. Exhaustive but ig with more $$s, comes more responsibilities !! That's it for the past 2 days, sorry for not tweeting, peace !!!
Siddharth tweet media
English
11
1
46
1.2K
Avid
Avid@Av1dlive·
Coding is going to be 100% done by AI by 2027 Every engineer is going to have and manage 100s of agents The future of work is managing and building ai agents [Here are 10 things senior agent engineers taught me on how to make Claude Code Agents:] →Skipping the 5 discovery questions before yapping at the agent: every run starts blank, every decision re-litigated. Blueprint → Links → Architecture → Style → Trigger, once →Telling the agent HOW instead of the OUTCOME: caps the run at your imagination. state the north star, let the model find a path you wouldn't →No decisions.md, no findings.md, no progress log: every cold start re-derives what last week already figured out. 5 markdown files beat 50 "as we discussed" messages →Re-wiring Firecrawl + Apify + Gmail + Calendar into every new project: same 4 MCPs, every repo, 20 min wasted. set them at the connector layer once, forget →Confusing the model with the IDE: Opus is the driver, Claude Code is the car. you've been tuning the car when the leverage is in the driver →Don't keep Swapping models mid-conversation: →Hand-tuning prompts in a feedback loop →decisions.md as the running ledger, not your short-term memory →skills for any flow you'd re-run more than twice →connectors set at the user layer, not per project →memory + loop + tools + context, written down, every project [what actually compounds instead: ] →cache-aware prompt layout (static first, dynamic last, never mutate the prefix) →state changes via messages, never via system-prompt edits →GEPA / DSPy on private-data tasks, vibes on the public ones →managed variables on anything you'd flip without a deploy →outcomes (rubric + grader) for subjective quality, deterministic evals for everything else →subagents for model switching, never a mid-chat swap →harness discipline over prompt cleverness in 12 months, the gap between teams shipping on Haiku and teams burning Opus on every turn won't be talent it'll be harness discipline study this.
darkzodchi@zodchiii

x.com/i/article/2054…

English
40
28
216
29.6K
Himanshu
Himanshu@nothiingf4·
good morning fellas :) - completed a night sprint of building 5-6 tools - the tools have around 23 features in total - checked and tested each of them - will be waiting for the reviews in my pr - will have to work on creating cookbooks and docs - i am about to pass out Discipline is not driven by environment ~ nothiingf4
Himanshu tweet media
English
1
0
22
270
Rohit
Rohit@rohit4verse·
this is how the founder of obsidian actually takes notes in his own app. most users get this wrong. barely any folders. heavy internal linking. categories live as properties on the note itself. article below teaches how to build personal knowledge base. video above is the underrated masterclass.
CyrilXBT@cyrilXBT

x.com/i/article/2053…

English
31
136
1.3K
208.1K
Avid
Avid@Av1dlive·
TODAY WE KILL EXCEL Claude Code handles the engineering side. the data ops side? still manual. still broken. still Excel. [ what Sanifu does for the other 80% ]: > finance team describes the process once in plain English : rules, exceptions, edge cases > Sanifu runs it end-to-end and shows real-time progress > they review, refine until it matches exactly > Sanifu saves it as a data app > every run since: same input + same rules = same output [ the part nobody's talking about ]: AI agents reinvent the logic every run.. you can't see it but the output shifts every time your reconciliation looks right.. but the methodology changed silently.. and you have no idea why fix: describe it once to Sanifu, lock the logic in [ and the craziest part ]: most teams automate the code. nobody automates the data underneath it. between you and full data ops automation there is one description and one approval most teams will never do this.. not because it's hard, but because they think it is try it: sanifu.ai
Sanifu.ai (YC S22)@SanifuAi

Today, we kill Excel Stop copy-pasting data, cleaning sheets, fixing formulas & working on the same excel every day. Describe the steps, rules and output once to Sanifu It executes it once, you publish & it runs the same way every time RT + comment "Sanifu", get 250k credits

English
6
2
16
2.1K
Avid
Avid@Av1dlive·
Anthropic CEO: "we got 7 more months." the bet was a $1B one-person company by end of 2026. This guy built the exact workflow that you need to make it happen Bookmark this and start it this week
CyrilXBT@cyrilXBT

x.com/i/article/2053…

English
29
13
109
24.6K
Himanshu retweetledi
Rohit
Rohit@rohit4verse·
mobile has been the broken half of AI app builders for a year. replit, lovable, bolt, cursor. ask any of them for an iOS app and you get a repo and a "good luck." runable hands you the app. installed on your phone. live in the App Store. live in the Play Store. tried explaining the difference to a friend. he didn't get it until he saw the icon on his home screen.
Umesh Kumar@itsumeshk

MacBook Pro. iPad. iPhone. Apple Watch. AirPods. All this is just the second prize. Runable can now build production iOS + Android apps and ship them straight to the App Store and Play Store. We're running a challenge, To participate: > Reply with your idea under this post > build a mobile app on Runable > record a quick demo > quote tweet this with your video + why you built it 2nd prize: the entire Apple ecosystem 1st prize: mystery starts today, ends sunday midnight.

English
12
4
43
9.2K
Avid
Avid@Av1dlive·
The person who built Claude Code just showed exactly how to build AI automations. 30 minutes. Free. Straight from Boris Cherny himself. Bookmark this before you forget. Most people using Claude daily are missing this feature hiding in plain sight. This single session is worth more than any $10,000 course.
Khairallah AL-Awady@eng_khairallah1

x.com/i/article/2053…

English
26
72
466
56.7K
Avid
Avid@Av1dlive·
OpenAI pays engineers $1,000,000+ a year to build agentic systems. Stanford just put a 2 hour lecture that covers 80% of it for FREE. Bookmark this. Give it 2 hours today. Then read the guide below It might be the highest ROI thing you do this month:
Avid@Av1dlive

x.com/i/article/2053…

English
17
52
240
21.5K
Utkarsh
Utkarsh@utk7arsh·
Hands are harder than Walking ⁉️🤔 Everyone in robotics is chasing bipedal locomotion. But the thing that will actually unlock humanoid robots in the real world is the hand. Last month, I listened from @tom_jiahao, founder of @DaxoRobotics, at upenn. He walked through the full landscape of robotic hand development. Three families of robotic hands exist right now. > Linkage drive hands are durable and expensive but not dexterous where motors sit in the palm. > Direct drive hands are fully dexterous but break often. > Tendon drive hands (Tesla hand, shadow hand) put the motors in the forearm, which is actually how human hands work: no bones in the hand, just tendons running from the forearm. Daxo's v1 squeezed 100+ motors together just to prove it could be done. Each finger now holds a decent payload. Cost is low. The hardware is solved. The bottleneck is data. A lot of simulation is replaced by imitation learning from real tasks, collected through wearable hand sensors. Getting that pipeline to scale is the open problem. The hand is next. Data is the hard part.
Utkarsh tweet media
English
5
0
18
540
Avid
Avid@Av1dlive·
"if you really want to make money, it's actually easy. found an agentic AI company." these are the exact words. by Eric Schimdt (ex-CEO of Google) [if I had 7 days to master agents and how to use them to automate workflows:] → read Anthropic: "Building Effective Agents" → read OpenAI: "A Practical Guide to Building Agents" → clone agentic-harness and run your first agent in under an hour that's the whole 7 days. bookmark this and start this week
Avid@Av1dlive

x.com/i/article/2053…

English
20
25
157
18.2K
Avid
Avid@Av1dlive·
Jensen Huang (CEO of Nvidia) : "AI is not likely to replace you, but someone using AI better than you might." two types of people right now: type 1: using AI to write emails. treating it like a search box. type 2:building skills. powering agents. sold it for $100,000. not get replaced I wrote the exact playbook for type 2. bookmark this. build this weekend ↓
Avid@Av1dlive

x.com/i/article/2053…

English
36
25
168
38.8K