Field guide / primary domain
40
sources in this field
Updated September 16, 2026
Current thesis
The shortest path to orientation.
This synthesis records claims and practices from the cited sources; reported outcomes and product capabilities have not been independently verified.
Screen-based knowledge work faces the highest automation exposure: Karpathy's BLS scoring (342 occupations, avg 5.3/10) identifies roles controlling $3.7 trillion in annual wages. Frontier agents dominate technical domains while general use-cases yield modest gains. Production evidence is aggressive—Marcus Moretti runs Spiral solo, Every scaled 4→30 employees while automating heavily, and Replit's 5.8x code increase translates to 2.9x per-engineer output after controlling for doubled headcount. The bottleneck shifted from typing to decision speed; human value concentrates on attention allocation and merge ownership. A ~$400/month multi-LLM stack delivers full dev-team capabilities, and developers are ~55% faster, yet no equivalent viral surge has appeared in writing—a sign that diffusion outside coding will take far longer than Silicon Valley expects. That slowness is exactly where applied-layer companies capture value, since coding diffused fast but most knowledge work hasn't. Income effects drive 75%+ structural shift toward high-elasticity sectors. Traditional CPO roles vanish within five years; the durable moat shifts to business context and accumulated workforce knowledge.
Evidence board
Claims worth carrying forward01Applied AI value lives in a layer that connects raw model intelligence to enterprise workflows: reengineering processes, aggregating context/data, designing human-in-the-loop experiences, driving change management, running domain-specific evals, and managing security/governance.
02As models get more capable, the applied/workflow layer becomes MORE important, not less—greater capability enables more complex tasks, which amplifies the cost of not building the integration layer well.
03Diffusion of AI outside of coding will take far longer than Silicon Valley expects; that slowness is exactly where applied-layer companies capture value, since coding diffused fast but most knowledge work hasn't.
04Box built a custom agentic harness that reportedly beats raw model API access on accuracy and latency for enterprise document tasks—supporting the thesis that harness/orchestration engineering outperforms bare model calls.
05Prediction: within five years, 90% of enterprise tokens will come from tasks that no human explicitly kicked off (i.e., autonomously triggered agent work), not human-initiated chat sessions.
06Letting model providers route your enterprise's tokens/traffic creates a 'fox guarding the henhouse' conflict of interest—application companies should be wary of ceding routing control to the labs whose models they depend on.
07Ian Vanagas draws a hard line between 'writing with AI' (using it for research, questions, and feedback while writing all prose himself) and 'using AI to write' (having it generate final text)—arguing the latter leaves 'skeletons of slop' even after heavy editing.
08He built a Claude 'researcher skill' to find real, quotable, sourced examples because unconstrained prompts caused hallucinated plausible-sounding examples; explicitly demanding sources and quotes keeps the model honest.
Adjacent fields
Latest evidence
Aaron Levie
Box CEO Aaron Levie argues that enterprise AI value comes from an 'applied layer' that closes the chasm between raw model capability and real workflows—via context aggregation, human-in-the-loop design, evals, and governance—and that this layer becomes more, not less, important as models improve.
Ian Vanagas
Ian Vanagas distinguishes 'writing with AI' (using AI for research, gap-finding, and fact-checking while keeping prose fully human) from 'using AI to write' (having it draft prose). He details a research skill stack for sourcing real examples, explains why AI is weak at summarization and editing/conciseness, and links this to a broader thesis on why AI hasn't produced 10x more 'banger' blog posts the way it has for code.
Shahaf Antwarg
A workflow synthesis from a developer running 20 parallel coding agent sessions: role shift to tech-lead, isolated worktrees per task, five-part first-message discipline, early direction-checking over end-diff review, manual commits, and a living instructions file.
ethan ding 📊
Ethan Ding's essay argues Palantir is a vertically-integrated outcome-based software provider whose moat comes from three compounding inputs: 12 years of patient capital during a commercial PMF desert, government deployments lacking modern cloud primitives, and an outcome-based (not usage or hours-based) pricing model. This combination produced a differentiated FDE talent pool, an unusually broad product (Foundry/Ontology), and account-level economics competitors structurally cannot replicate, making Palantir near-impossible to clone in the current AI/VC funding environment that demands 3-month results.
Guillermo Rauch
Guillermo Rauch describes Vercel's internal agent @v, which handles finance, comms, docs, marketing, engineering, and business analytics across the company, growing exponentially in usage. It maintains per-user memory and personalized scheduled workflows, and directly informed the design of evedev.
Sierra
Sierra built Pinecone, an internal cloud agent platform that unifies employee AI usage into a single shared, durable, multiplayer system rather than individual laptop-based agent sessions. The design philosophy centers on building durable primitives (context, environments, tools) rather than workflows, centralizing improvements company-wide, and treating accumulated workforce knowledge as the key competitive moat as frontier model intelligence commoditizes.
Amjad Masad
Replit's internal post details how deeply integrated agents (via Slack, GitHub, GCP, Linear, Notion, ZenDesk) tripled per-engineer code output while holding review time, quality, and incident rates flat, then spread beyond engineering into sales, marketing, support, and data teams — reframing AI adoption as promotion rather than automation.
George from 🕹prodmgmt.world
A framework for writing decision memos that combines top-down (principles-to-specifics) and bottom-up (evidence-to-conclusions) analysis, comparing them on one page to surface agreement, conflict, and gaps before stating a conclusion-first recommendation.
PULKIT WALIA
OpenAI pays $280K for Forward Deployed Engineers (FDEs). The hiring process is explicitly not LeetCode-style, implying a different interview loop focused on practical deployment and customer-facing AI work. No further detail on the actual loop is provided in the post content.
Jason Zook
Jason Zook shares a two-model software development workflow using Claude Opus 4.7 for planning and building, GPT-5.5 for adversarial review, and Playwright for automated UX/UI testing. The loop runs plan→review→update→build→review→fix→deploy, with ~$400/month replacing what feels like a full dev team.