Proby Shandilya

0 followers · 306 views

on the atlas — 37

highlights — 56

If there is enough margin of safety, we don’t have to worry about the short-term fluctuations of the market at all.
The Uncommon Sense of Li Lu's Partner, Jing (Gene) Chang
a system that can surpass humans in all intellectual activities and design better machines to improve itself
Harness Engineering for Self-Improvement | Lil'Log
We estimate that $240 billion to $280 billion in infrastructure capital expenditures across eight key sectors enter the federal permitting process each year.
How permitting will impact US infrastructure projects | McKinsey
gap grows between the country’s existing infrastructure and what is needed
How permitting will impact US infrastructure projects | McKinsey
Natural ecosystems are invaluable both in their own right and in the services they provide to societies—for example, clean air, clean water, quality soils, climatic regulation, and recreation.
How permitting will impact US infrastructure projects | McKinsey
It is a new collaborative surface for building software in the open inside Harvey.
Building Spectre: Internal Collaborative Cloud Agent Platform | Harvey
built a central internal MCP server called Toolshed, which hosts more than 400 MCP tools spanning internal systems and SaaS platforms we use at Stripe
Minions: Stripe’s one-shot, end-to-end coding agents | Stripe Dot Dev Blog
minion creates a branch, pushes it to CI, and prepares a pull request following Stripe’s PR template
Minions: Stripe’s one-shot, end-to-end coding agents | Stripe Dot Dev Blog
While we provide CLI and web interfaces for initiating minions, engineers will most frequently start one from Slack. By tagging our Slack app, engineers can kick off a minion directly from the thread discussing a change, and it’ll be able to access the entire thread and any links included as context.
Minions: Stripe’s one-shot, end-to-end coding agents | Stripe Dot Dev Blog
Minions use the same developer tooling that equally enables Stripe’s human engineers to effectively operate on our scale: if it’s good for humans, it’s good for LLMs, too.
Minions: Stripe’s one-shot, end-to-end coding agents | Stripe Dot Dev Blog
However, iterating on any codebase of the scale, complexity, and maturity of Stripe’s is inherently much harder.
Minions: Stripe’s one-shot, end-to-end coding agents | Stripe Dot Dev Blog
unattended agents allow for parallelization of tasks.
Minions: Stripe’s one-shot, end-to-end coding agents | Stripe Dot Dev Blog
Over a thousand pull requests merged each week at Stripe are completely minion-produced, and while they’re human-reviewed, they contain no human-written code.
Minions: Stripe’s one-shot, end-to-end coding agents | Stripe Dot Dev Blog
Each session runs in a sandboxed VM on Modal with everything an engineer would have locally: Vite, Postgres, Temporal, the works. It’s wired into Sentry, Datadog, LaunchDarkly, Braintrust, GitHub, Slack,
Why We Built Our Own Background Agent — Ramp Builders Blog
~30% of all pull requests merged to our frontend and backend repos are written by Inspect. It only took a couple months for us to reach this level of usage, and it continues to grow.
Why We Built Our Own Background Agent — Ramp Builders Blog
When background agents are fast, they’re strictly better than local: same intelligence, more power, and unlimited concurrency. You can go home and let Inspect cook (and if you’re so inclined, resume after dinner from your couch and mobile phone).
Why We Built Our Own Background Agent — Ramp Builders Blog
session speed should only be limited by model-provider time-to-first-token
Why We Built Our Own Background Agent — Ramp Builders Blog
First, a patient’s cells are extracted, frozen, and transferred to a central manufacturing facility, where they are engineered to express the CAR-T receptors and screened for quality. Next, they are frozen again and transferred back to the hospital where they need to be administered.
Medicine's Endgame
This has primarily been the byproduct of the industrialization of synthetic chemistry, which gave us small molecule drugs like statins that lower LDL cholesterol levels, reducing the risk of heart disease for millions of people around the world.
Medicine's Endgame
the gap between what models know and what they could know has become increasingly obvious.
Why We Need Continual Learning | Andreessen Horowitz
Conversely, if we can train models to learn their own memory architectures – rather than offloading to bespoke harnesses – we may unlock a new dimension of scaling.
Why We Need Continual Learning | Andreessen Horowitz
The agent lacked the tools, abstractions, and internal structure required to make progress toward high-level goals. The primary job of our engineering team became enabling the agents to do useful work.
Harness engineering: leveraging Codex in an agent-first world | OpenAI
introduced a different kind of engineering work, focused on systems, scaffolding, and leverage.
Harness engineering: leveraging Codex in an agent-first world | OpenAI
We estimate that we built this in about 1/10th the time it would have taken to write the code by hand.
Harness engineering: leveraging Codex in an agent-first world | OpenAI
They are equally kind, deep, and competitive but present very differently: Varun pushes, runs hot, moves fast, sweats the details, and can get obsessed with a particular problem or person. Kareem is more airy, philosophical, and often gives space when you expect direction. On a fundraise announcement morning, Varun is editing commas at midnight and in the conference room at 7AM; Kareem has to be mildly bullied into reading the materials and rolls in well after press is live.
Firsthand: How I Bet On Clay (And It Bet On Me)
Kareem just made me feel like we were the same kind of person.
Firsthand: How I Bet On Clay (And It Bet On Me)
He was the main reason Clay hired so well and fast early on, and he still works the same way: he finds people that spike at something, gets them in a room as fast as possible, and has already intuited where they might fit by the time they sit down.
Firsthand: How I Bet On Clay (And It Bet On Me)
You have to balance three forces: what customers want, what's best for productization of your services, and how hard the product element is to build. The founders who get this right learn to say no to customer requests that don't map to their productization roadmap, even when the revenue is tempting.
The AI-Native Services Playbook | Emergence Capital
Customer demands are loud and immediate. Internal platform investments are quiet and compounding. The failure mode is to become a traditional services business by neglecting platform development while responding to urgent client requests.
The AI-Native Services Playbook | Emergence Capital
tooling used to design any physical product from semiconductors, to rockets, to toothbrushes.
Physical Design Software: The Strongest Moats in Software
rise of specialized AI agents that work alongside engineers throughout the product lifecycle
Agentic Engineering: how AI automata will participate in engineering in 2025 - Blake Courter
these inventions and discoveries don’t exist in isolation, they bleed into each other and are remixed in new ways to accelerate progress.
Medicine's Endgame
His daughter had recently been prescribed a medication targeting IL6 called Tocilizumab to manage her case of juvenile rheumatoid arthritis. June had a long-shot idea: what if they dosed Emily with Tocilizumab to curb the inflammation?
Medicine's Endgame
Far from Coley’s brilliant but crude “toxin” injections, Eshhaar was equipped with an entirely different language for how the body’s immune system battles disease.
Medicine's Endgame
Through a profoundly complex cascade of genetically controlled cellular divisions, we ditch our microbial ancestors and grow into a mosaic of cells that are hierarchically organized into tissues and organs.
Medicine's Endgame
By engineering and reprogramming our own cells, we may be able to cure cancer, manufacture drugs inside our body, regenerate organs, and even reverse aging.
Medicine's Endgame
translating requests into prompts, monitoring agents, and merging PRs
Engineering as a Service
we're also selling you a style of working and a system on which you do your work.
Building and Investing in Lattice - Colossus
That we don't just sell you a product, we're selling you on being the kind of person who wants to buy this kind of product.
Building and Investing in Lattice - Colossus
And what people are really buying is -- I think about it as a way of being, like a way of achieving success.
Building and Investing in Lattice - Colossus
existing NHI management players are well positioned to respond to the risks
Authenticating AI Agents — Activant
These companies also offer NHI lifecycle management and offboarding tools – so AI agents’ privileges can be revoked when they are no longer needed.
Authenticating AI Agents — Activant
NHIs are growing at a rapid rate of ~2.5x every year, outnumbering human identities by a factor of 45x.2 Yet, the amount spent on NHI security is only 5%–10% of the $76bn spent on human identity security
Authenticating AI Agents — Activant
Fifty-five percent of organizations have experienced a breach or incident via their software supply chain that exploited vulnerabilities in NHIs.1
Authenticating AI Agents — Activant
energy state, an attitude, and an unexamined philosophy
The Resourceful Life
RAG is great for document search, but our insurance workflows already have a fixed set of inputs.
Pace | Building Reliable Insurance Agents
successful agents in production mirror how human experts actually work
Pace | Building Reliable Insurance Agents
The core challenge of long-running agents is that they must work in discrete sessions, and each new session begins with no memory of what came before.
Effective harnesses for long-running agents \ Anthropic
nto $40 billion. His investments returned 19% annually. He beat the index by 5% a year.
The Last Human Edge - Colossus
he turned $8 billion into $40 billion.
The Last Human Edge - Colossus