running a 15 person growth company, i was drowning in tools that refused to talk to each other.
my mornings looked like this:
• open jibble to see who clocked in
• open an excel with 306 tasks across 6 tabs
• scroll the chat group for everyone's "today i will work on" messages
• dig through a google form sheet for a new joiner's PAN
• open one sheet for salaries, another two for brand logins
every answer lived in a different tab.
every approval waited on me.
every exit meant remembering to revoke access in 6 places.
so I built NORA. our own internal HRMS and ops app. 16 days of building, runs on firebase, $0/month infra.
what's inside:
• one task board for the company. assign to a person, a department, or everyone. one click moves a task onto your day plan.
• a time clock that understands timezones. default shift is 12-9 IST, our SDR runs 8 PM-2 AM for US hours, and her 1 AM clock-out files on the right day. forgotten days close themselves at night.
• day plans replaced the daily chat messages. morning bullets, tick them through the day, the ticked list is the evening report. the whole team sees everyone's.
• leave that runs itself. balances accrue by policy, requests climb an approval ladder, an admin's request comes only to me. wrong approval? revoke puts the days back.
• a payroll register the CA can use directly. we compute gross, TDS and PT per person, prorate for joiners and exits, and hand off one CSV. the P&L pulls from it live.
• a vault for shared logins. masked passwords, copy without revealing, admin-only master shelf. both access sheets retired.
• the brain: a versioned library for our AI skill files and every brand's context docs. anyone uploads, every update becomes v(n+1), and we keep the history. skills stopped travelling by chat.
• onboarding collects PAN, aadhaar and bank details in-app. offboarding is one click and actually revokes access, session tokens included.
• an org chart, notices, a team wall for birthdays and wins, a feature request board with votes.
• a 10 AM digest email to each person: overdue, due today, blocked, upcoming. founders ride CC on every send.
Helped me tremendously as a founder:
• 6 tools and sheets became 1 url
• approvals happen where the data lives. i decide in one tap and the balance updates itself.
• payroll prep dropped from an evening to minutes
• i see the whole company's day without pinging a single person
• 900+ automated tests and an audit trail sit behind it, so i trust it more than i ever trusted our sheets
a 15 person team runs fine on one app shaped around how it actually works. And the best part - Ill prob be able to run this for under $20 a month as it scales to 100 folks
@jun_song The obsession needs to be studied too. $200 gives a frontier model, regular updates , and subsidized to extract 10X value than the price paid,
how many can justify a real use case for local llm .
I agree that handcrafted feels better. But when I first looked at the ad , it felt done by AI, and I was still impressed.
Not sure something done by AI automatically triggers a negative emotion for many.
You can now set Claude Code's output style to Concise.
Claude leads with the result, keeps responses short, and still gives full detail when you ask.
Turn it on in /config → Output style, or set "outputStyle": "Concise" in settings.json.
organic search traffic in aggregate is down about 2.5% year over year, basically flat.
but organic click through rate drops 61% on any query where an AI overview shows up, and if your brand is the one actually cited inside that overview, your click through rate runs about 35% higher than normal.
the traffic moved from the ten blue links to whoever gets cited inside the AI summary.
the people yelling seo is dead are looking at the flat aggregate number. the people yelling seo just changed are looking at the citation data. they're both staring at the same dataset from a different altitude.
the actual job now is earning the citation inside the answer. ranking position one through ten stopped being the point.
that's a different skill. it runs on structure and clarity, and keyword density barely moves it anymore.
google folds automatically created assets and campaign level broad match into AI Max on september 1st.
two weeks out.
the last manual levers most advertisers had over ad copy and match type inside a standard campaign disappear into one automated layer you can't fully see inside.
i've watched accounts lose 15-20% efficiency in the first month after a forced migration like this.
the ones that got hurt skipped one step: the team never pulled a baseline of what was actually running before the switch. if you have live campaigns, this is not the week to skim past an email from google.
pull your current asset performance, flag what's actually converting, know exactly what you're losing visibility into before the 1st.
after that date you're reading results without being able to see the causes behind them.
your competitor isn't smarter about AI than you. they just answer the phone faster and follow up more times than you do.
here's exactly how they set that up, week by week, in 90 days. https://t.co/bo1mcanvv8
We asked an unreleased research version of Claude to take a stab at the Riemann hypothesis.
It didn’t solve it, but it did make strides on a related problem: it increased the lower bound for the fraction of zeros of the Riemann zeta function that satisfy the hypothesis from 41.6% to 67.2%.
https://t.co/aZDvqqhHRi
New in Claude Code: your sessions can now message each other.
Instead of having to re-explain yourself in another session, you can now tell Claude to do it. It sends a summary (not your history or files), and the other session picks it up mid-task.