The Unread Pile

Monday, 28 September 2026 · MumbaiEdition No. 425 new posts · 26–28 Sep

Lead Feature · Five saves, one thesis

The Agent Engineering Playbook

Spine: Andrej Karpathy’s free 2-hour lecture, via @impreiaxbt · Supporting saves: FarzaTV ×2, @0xwhrrari, @championswimmer

Five of this week’s saves say the same thing from five angles: the agent era has stopped being about prompts and started being about engineering. Karpathy released a free two-hour lecture on agents — loops, the harness, self-improving systems — and the rest of your pile fills in the discipline around it: how to build the environment that makes agents work, how to stop them bloating your codebase, and where the economic value actually lands. The prompt is the demo. The loop is the product.

Working late by lamplight — The loopy era: Karpathy reportedly runs his own autoresearch loop sixteen hours a day. The lecture is the theory; the overnight run is the practice.

Karpathy’s lecture arrives at the moment his own experiment went from demo to discipline. His open-source autoresearch project (March 2026) put an agent in charge of improving single-GPU LLM training: the agent edits train.py, runs a fixed five-minute experiment, scores validation bits-per-byte, keeps what improves, and iterates. Seven hundred experiments over two nights, twenty optimisations discovered — greedy hill-climbing over code edits with a fixed compute budget so every run is comparable. The loop, in its cleanest form: Literature → Hypothesis → Code → Execute → Analyse → Iterate. Researchers have since shown the gains come from loop structure, not model capability — the same model at the meta level, five times the improvement, because the outer loop rewrites search code instead of tuning parameters.

The autoresearch loop — the cleanest agent loop in the wild

After Karpathy’s autoresearch (Mar 2026); loop mapping via the adev-plugin research survey.

Your other saves are the operating manual for that loop. @0xwhrrari frames it as harness engineering: the job is to build the environment — sandbox, tools, a fixed clock, an honest fitness function — that makes the agent’s iterations comparable and its wins cumulative. @championswimmer supplies the sharpest negative lesson: prompt skills and rules files cannot stop code bloat, because the agent that wrote the mess can no longer see it. The fix is a fresh-context review agent — a second pass with clean eyes, the way a senior dev reviews a junior’s PR. And Karpathy’s own coding workflow, broken down this month, distils to three guardrails your saves keep circling: think before coding (state assumptions before touching the codebase), simplicity first (if fifty lines solve it, never let the agent write two hundred), and surgical changes (no drive-by refactors). The pattern across all of them: the agent behaves like a hasty, over-confident junior developer, so you manage it like one — with process, not hope.

The harness — what actually makes the loop work

Synthesised from @0xwhrrari (harness engineering), @championswimmer (fresh-context review), and the Karpathy workflow breakdown.

Then FarzaTV answers the “so what” twice. First, the killer use-case for personal agents: not answering questions, but helping individuals create economic value — agents as leverage for one person’s output, which is the demand story for everything above. Second, the product shape that follows: collaborators, mini-AIs that help you build, find users, and earn your first dollar. It’s the same arc as the Harsha Kumar essay from Edition 3 — the assistant that acts — now with an engineering manual attached.

“Stop your AI coding agent from acting like a junior developer.”— Andrej Karpathy, via the workflow breakdown of his agentic coding setup

700

experiments in 2 nights (autoresearch)

2 hrs

free Karpathy lecture on agents

16 hrs

a day Karpathy runs the loop

5×

gain from loop structure, not model size

Main learnings

The loop is the product. Comparable experiments (fixed budget, honest metric) plus a learnings store beat a smarter model every time.
Manage the agent like a junior dev. State assumptions, demand simplicity, forbid drive-by refactors — in process, not in prompts.
Fresh context beats more context. A second agent with clean eyes catches bloat the author-agent is blind to.
Economic value is the killer use-case. Personal agents win by making one person produce more, not by chatting better.

Practical takeaway — Before your next agent session: set a fixed time/compute budget per iteration, define the one metric that decides “keep or discard”, and schedule a fresh-context review pass at the end. That’s the whole harness.

Originals: Karpathy’s 2-hour lecture · FarzaTV: collaborators · FarzaTV: agents & economic value · 0xwhrrari: harness engineering · championswimmer: fresh-context review

The Capability Gap

Capability is sprinting. Adoption is walking.

Via @akothari · Pat Grady’s slide · plus @waiting4_asi and @PromptLLM

@akothari shared a Pat Grady slide making the cleanest point in this edition: the gap between what AI can do and what organisations have adopted is enormous — and gaps that big are where fortunes are made. @waiting4_asi puts a clock on it: a three-to-four-month AI arbitrage window, in which whoever exploits the niches first wins fast. And @PromptLLM shows the frontier moving again — an image post on how Opus 5.5 would beat everyone else — which only widens the gap further.

The gap — capability vs adoption, 2024 → 2026

After Pat Grady’s slide, via @akothari — the arbitrage window from @waiting4_asi.

The mechanism is simple: capability compounds on research timelines while adoption compounds on organisational timelines — budgets, habits, procurement. Every model release re-opens the spread. The arbitrage window is the period between “the model can do it” and “everyone’s procurement approved it”. In that window, a small team with no approval process can ship what an enterprise is still forming a committee about. Three to four months sounds short; it’s roughly how long the gap has been open, on repeat, for two years.

Main learnings

Watch the gap, not the leaderboard. The money is in the spread between capability and adoption, which every release refreshes.
Speed is the moat inside the window. Small teams win arbitrage plays because they skip the committee.
Niches first. The window rewards whoever picks an unguarded niche and exploits it fast, not whoever builds the biggest model.

Practical takeaway — Ask, for your own work: what became possible in the last two model releases that your org hasn’t adopted yet? That list is your next quarter’s roadmap.

Originals: Pat Grady’s slide · the 3–4 month arbitrage window · how Opus 5.5 would beat everyone

Business · Private Equity, Rewritten

The $5 Trillion AI Roll-Up

By @gregisenberg · 27 Sep

Greg Isenberg’s thesis is the capability gap, weaponised: buy boring businesses, then triple their EBITDA with agents. The private-equity roll-up playbook — acquire at low multiples, centralise, cut costs — has always worked; AI agents just made the “cut costs” step an order of magnitude more violent, and available to anyone with a laptop instead of a fund.

Product and operations — The roll-up target: unglamorous, cash-flowing, and barely digitised — exactly the businesses agents transform fastest.

The roll-up math — one boring business, with and without agents

Illustrative math on Isenberg’s thesis: acquire at ~3× EBITDA, triple earnings with agents.

Why this is a $5T idea and not a $5B one: the target set is every service business that never got the software treatment — the plumbers’ dispatchers, the clinics’ schedulers, the logistics brokers. Software ate the front office; agents are eating the back office. And the buyer doesn’t need to be a fund anymore: the harness from this edition’s lead story is the operating partner.

Practical takeaway — If you can run the agent loop from the lead story, you can run the diligence: pick one boring local business, map its three most repetitive workflows, and price what agents do to its margins. That’s the pitch deck.

Original: $5T AI roll-ups: buy boring businesses, 3× EBITDA with agents

Notes on Sequoia’s Playbook

Via @0xTykoo · 26 Sep

Notes circulating on how Sequoia thinks — the firm that backed Shaadi.com back in 2006 and has ridden every one of these cycles. Read it as the institutional version of the arbitrage thesis: the best firms don’t predict the gap, they position inside it early and let the compounding do the work.

Original: Notes on Sequoia’s…

The AI Update, Boardroom Edition

Via @gradypb · 25 Sep

A Loom recording of an AI update delivered to the Boston College investment committee — the adoption side of the capability gap, in the wild. Endowments move slowly; the fact that “AI update” is now standing agenda at investment committees tells you the gap is closing from the demand side too.

Original: Loom: AI update for the Boston College investment committee

Organisations

How to Level Up an Org, in Three Phases

By @spakhm (Slava Akhmechet) · 10 Sep · plus @vijayshekhar

A person at work — Phase three is the whole game: the stars were already there, hidden by the org chart.

Slava Akhmechet’s framework for levelling up organisations runs in three phases — learning, changes, elevation. Phase one is pure observation: learn how the org actually works, who really decides, where the energy is. Phase two is making the changes the learning revealed — structure, process, the hard calls. Phase three is the payoff most leaders skip: elevate the hidden stars, the people doing the real work two levels below where the credit lands. The framework rhymes with the agent loop from the lead: observe (literature), act (code), keep what works (analyse) — management as hill-climbing.

The three phases of levelling up an org

After @spakhm’s post on levelling up organisations.

@vijayshekhar adds the cultural footnote: CEOs speaking their mother tongue — the signal that leadership has stopped performing for someone else’s template. It reads as the human version of phase three: authenticity as elevation, the org finally speaking in its own voice.

Practical takeaway — New to a team or leading one? Spend your first month in pure learning mode, write down the three hidden stars, and make elevating them your first visible act.

Originals: how to level up orgs · CEOs speaking mother tongue

Founder Story · Thirty Years On

The Shaadi.com Origin Nobody Remembers

Thread via @Rahul_J_Mathur · the story is Manish Jain’s · 25 Sep

Before it was Anupam Mittal’s sagaai.com, shaadi.com was two guys, a brand-new internet, and a domain bought from Network Solutions — the only place you could buy one. Manish Jain’s thirtieth-anniversary post is the best founder story in this edition: no grand plan, no market research, just a name that felt obvious and a static site built in Microsoft FrontPage.

Writing the first version — Version one was a static site in FrontPage. It still worked.

9 Aug 1995 · The spark

The Netscape IPO.

Manish Jain and his brother-in-law Sandeep Jain kick around the two things they believe the internet will transform — jobs and matrimonials — and pick matrimonials.

Jul 1996 · The domain

shaadi.com bought from Network Solutions.

No registrars, no auctions. One store, one price.

1 Oct 1996 · Launch

A static FrontPage site goes live.

The response, in Jain’s words, was “beyond our wildest expectations.”

Late 1996 · The Hotmail accident

3,982 members — 35% of the database — sign up with Hotmail addresses.

Hotmail itself was 89 days old at launch. Two things that didn’t exist at the start of the year carried more than a third of the members.

1997–1999 · The other origin

Anupam Mittal launches sagaai.com (1997), renames it shaadi.com (1999).

His spark: a traditional matchmaker posing as his father’s friend, trying to size him up — “we could put all this on the Internet.” No external capital until Sequoia’s round in 2006.

2000 · The exit

Jain sells the original site.

“You had the wrong dream,” he quotes — everything after was timing, luck, and people braver than them.

“Honestly, we had no idea what it would become.”— Manish Jain, on buying shaadi.com thirty years ago

Main learnings

Obvious names are underpriced. “A name that felt obvious” beat every branding exercise — twice, by two different founders.
Timing is a feature. The Hotmail accident — 35% of members arriving on an 89-day-old email service — is the whole 1996 internet in one stat.
Bootstrap until the brand exists. Mittal waited nine years and a real business before taking Sequoia’s money in 2006.

Practical takeaway — The best origin stories share a shape: a sharp observation (the matchmaker, the Netscape IPO), an obvious name, and a version one so simple it’s embarrassing. Ship that.

Original: thread on Shaadi.com’s origin · Manish Jain’s anniversary post: celestri.org

Claude-Made Cinema

Paul Graham’s Best Essay, in 200 Seconds

By @deedydas · two saves, 27 Sep

Deedy Das keeps proving the same point: a single essay plus Claude is now a film studio. This week’s exhibits — Paul Graham’s “How to Do Great Work” compressed into a 200-second AI video, and the entire history of Google told in two minutes, both generated with Claude.

An essay becomes a film — One essay in, one film out: the essay-to-video pipeline is now a weekend project.

The essay-to-video pipeline

The pipeline behind @deedydas’s AI-generated explainer videos.

The deeper point: the binding constraint on educational media was never cameras or editing — it was the compression of a long text into a tight visual script, and narration, and pacing. All of that is now one model’s afternoon. Deedy’s fully-automated YouTuber pipeline from Edition 3 was the factory; these two videos are the product line running. Expect the “200-second version” of every canonical essay within a year — and expect the winners to be the ones with taste in what to compress, not access to the tools.

Practical takeaway — Have a canonical text your team should have read but hasn’t? Commission the 200-second version. The cost is now a prompt, not a production crew.

Originals: PG’s “How to Do Great Work” as a 200-second video · Google’s history in 2 minutes, via Claude

India, Tested

Tata Tests Where It Matters

Via @arpit_vermaniac · 25 Sep

Tata’s ad — cars tested in real Indian conditions — went viral, and the reason is structural: Indian roads are the world’s hardest durability lab. @nitprashant makes the companion point: India’s worst traffic doubles as the ultimate car durability test. Marketing that shows the actual test, on the actual roads, beats a studio shoot because the audience drives those roads.

Practical takeaway — If your product survives your hardest real-world conditions, show the test, not the studio. Adversity is the most credible ad format there is.

Originals: Tata’s viral real-conditions ad · traffic as a durability test

The Mother-Tongue CEO

Via @vijayshekhar · 26 Sep

Filed under organisations above, but worth its own beat: the Indian founder who stops presenting in borrowed language. As India’s startup voice matures, the mother tongue isn’t folksy — it’s precision. You argue best in the language you think in.

Original: CEOs speaking mother tongue

The Briefing Room

From scarcity to abundance @arnav_kumar · 26 Sep
A generational mindset shift: the scarcity programming of the previous decade giving way to abundance thinking. Pairs well with the arbitrage window above — abundance mindsets are what let you act inside it.

DHH’s talk will prove prophetic @sriramk · 26 Sep
“Every engineer should hear DHH’s talk” — flagged as the kind of talk that reads obvious now and prophetic later. The DHH corpus (Rails, remote work, now post-framework thinking) has a habit of being early.

Top 1% speaker, in one page @gregisenberg · 25 Sep
A one-page distillation of what separates top-percentile speakers. The through-line with the rest of the edition: compression is the skill — essays to 200 seconds, talks to one page, orgs to three phases.

A mantra for procrastination @PromptLLM · 26 Sep
A simple mantra to cure procrastination. File next to Karpathy’s guardrails: most productivity systems fail at the starting step, and a mantra is a starting ritual with zero setup cost.

The cute-companion robot @MohapatraHemant · 26 Sep
Robotics form factor going soft: cute, evolving companions rather than industrial arms, with an eyecandy robotics launch attached. The form factor is the go-to-market — nobody fears a companion.

Revenge, automated @shikhars_ · 24 Sep
A joke prompt with a serious substrate: automating revenge on WhatsApp promo spam. The funniest save in the pile — and a reminder that the agent loop works on annoyances too.

Slava’s AGI test @spakhm · Aug
“When AGI finally arrives it will be blindingly obvious and there will be no debate whether it’s here — so long as reasonable people earnestly disagree, it is not here.” The cleanest falsifiable test in the discourse.

How Opus 5.5 would win @PromptLLM · 27 Sep
An image post mapping how the next Claude would beat everyone else. Read as the next tick of the capability line in the gap chart above.

Main learnings

In this edition · 25 posts

Main learnings

Main learnings