Tweet
Mar 26, 2026
AI at Work
Tweets, posts, and field notes on how people and companies are actually adopting AI at work.
-
-
Tweet
Mar 26, 2026
What If AI For Work Was Just Like TikTok's For You Page?
Greze makes a product argument rather than a model argument: the next wave of workplace AI may look more like a recommendation engine than a chatbot. The core claim is that adoption improves when AI handles obvious tasks proactively, then uses chat as a follow-up interface instead of the starting point.
-
Tweet
Mar 25, 2026
Durable advantages in AI products and agent systems
The post separates fast-decaying technical edges from slower-moving product and operational advantages. It is a useful reminder that durable leverage may come less from squeezing model benchmarks and more from UX, integrations, and faster agent feedback and verification loops.
-
Tweet
Mar 25, 2026
Three layers in the agent commerce stack
The post breaks agent commerce into three businesses: packaging tools and workflows as callable services, the payment and billing rails behind them, and discovery systems that help agents choose among competing endpoints. The key claim is that ranking, reputation, and trust will become especially valuable as the number of machine-consumable services grows.
-
Tweet
Mar 24, 2026
Entering the Era of the Headless Merchant
Levine highlights a shift from human-facing checkout flows toward services that expose themselves as machine-consumable products. The underlying idea is that agent commerce becomes more practical when merchants publish APIs, pricing, and workflows that software can buy and use directly.
-
Tweet
Mar 23, 2026
How we made Ramp Sheets self-maintaining
Ramp Labs outlines a practical agent workflow for operating production software: monitor live systems, surface issues, and prepare fixes while keeping engineers in the approval loop. It is a useful example of AI applied to maintenance work rather than greenfield generation.
-
Tweet
Mar 23, 2026
The Agentic Economy Will Be Massive. Agentic Commerce Won't
A useful counterpoint to the loudest agentic commerce narratives. The core claim is that most economic activity from agents will likely be bundled into existing software contracts or routed through humans at the point of choice, which makes autonomous machine-to-machine payments a narrower category than the hype suggests.
-
Tweet
Mar 23, 2026
AI demands a new proof of work
Rampell frames AI spam as an economic problem more than a filtering problem. The core idea is that if AI makes personalized outreach effectively free, work channels may need some form of proof of work or paid access to stay usable.
-
Tweet
Mar 22, 2026
Pirates and architects
A concise org-design frame for AI-assisted product development: one role pushes quickly toward shipped value, while another consolidates the resulting surface into a more structured system. The post is useful for teams thinking about how coding agents may change staffing patterns, ownership, and the timing of architecture work.
-
Tweet
Mar 22, 2026
Recursive AI will not preserve strategic advantage
A concise argument against assuming enterprise AI creates durable strategic compounding by default. The post points to two frictions that matter in practice: people protect their undocumented edge, and once AI systems become widespread, their outputs can start to look alike.
-
Tweet
Mar 22, 2026
Who wins agentic commerce?
A useful strategy lens on agentic commerce: the important question is not whether AI-assisted checkout appears, but which layer captures durable value once these flows become widely available. The post is relevant for teams thinking through platform power, distribution, and infrastructure in AI-native commerce.
-
Tweet
Mar 20, 2026
Open Agentic Commerce
Ragsdale argues that agentic commerce only becomes transformative when agents can buy from the open internet without pre-approved merchant relationships. The core claim is that open payment and discovery rails could do for agent-driven work what open web protocols did for the early internet.
-
Tweet
Mar 20, 2026
AI compute budgets will move beyond engineering
The post frames AI adoption as a budgeting and operating model question, not just a tooling decision for engineering. As agents take on more token-intensive work across functions, companies will likely need new ways to allocate and manage compute spend outside a traditional IT budget.
-
Tweet
Mar 18, 2026
AI will raise the premium on taste and judgment
If execution becomes abundant, discernment becomes more valuable. Teams still need people who can define quality, make tradeoffs, and recognize what should ship.
-
Tweet
Mar 17, 2026
Agentic software work needs ruthless deletion, not just better tools
As software gets cheaper to generate, teams need stronger editing and deletion habits. The bottleneck shifts from producing code to curating what deserves to remain.
-
Blog post
Mar 17, 2026
AI fatigue comes from review overload and constant tool churn
Khare argues that AI shifts engineers from making to reviewing, which can increase fatigue even when task-level throughput improves. Faster generation does not remove the human cost of coordination, judgment, and context switching.
-
Tweet
Mar 17, 2026
Validation, not code review, is the real agentic engineering bottleneck
Faster code generation does not remove the need for trust-building. As more changes are agent-produced, validation capacity across QA, security, and operations becomes the scarcer resource.
-
Tweet
Mar 16, 2026
Authentic human charisma becomes more valuable as AI mediates more work
As synthetic fluency becomes cheaper, distinctive human presence may become more valuable by contrast. The advantage shifts from polished output alone to unmistakable personality and point of view.
-
Tweet
Mar 16, 2026
Multi-agent systems will feel more ecological than mechanical
As agents interact across open systems, predictability drops and ecosystem thinking matters more. Teams may need to manage incentives, interfaces, and containment rather than assume a single deterministic workflow.
-
Tweet
Mar 16, 2026
Uber's internal coding agent shows agentic engineering is scaling
This is a practical signal that agentic coding is moving from demo territory into measurable production throughput. Once that happens, planning, review, and validation become the harder scaling problems.
-
Blog post
Mar 16, 2026
AI may push white-collar work toward a trade model
The essay argues that AI may change not just headcount but the social contract around white-collar work. As cognition gets cheaper, autonomy and strategic influence may concentrate in fewer roles while more jobs are scoped around execution.
-
Tweet
Mar 16, 2026
LLMs may speed startup work while hurting long-term velocity
Short-term speed and long-term velocity are not the same thing. If code volume grows faster than review and maintenance capacity, the cost appears later as complexity and lower confidence in change.
-
Tweet
Mar 15, 2026
The magic of AI products is collapsing browser tabs and spreadsheets
The value here is interface compression, not just automation. When AI replaces glue work across tabs, tools, and trackers, the workflow feels more native and requires less coordination overhead.
-
Tweet
Mar 15, 2026
AI is likely to complement many susceptible jobs, not just substitute them
This is a useful reminder that exposure is not the same as displacement. In many roles, AI can increase the value of workers who direct, verify, and compound the tool effectively.
-
Tweet
Mar 15, 2026
Slop creep happens when teams hand thinking to coding agents
The risk is often gradual quality erosion rather than one dramatic failure. Teams still need architectural judgment and explicit quality bars so agent output compounds into coherence instead of drift.
-
Blog post
Mar 15, 2026
Slop creep is the slow enshittification of a codebase
Boris argues that agent leverage raises the cost of vague thinking. When agents make many locally reasonable decisions without strong direction, quality can decay faster than normal review loops catch it.
-
Blog post
Mar 14, 2026
The production function changed
Strong framing for AI at work: software output is no longer primarily constrained by typing speed or manual implementation. If this holds, teams should optimize less for coding throughput and more for sharper specs, tighter feedback loops, and robust review guardrails.
-
Tweet
Mar 13, 2026
Management In The Age Of AI
This frames AI adoption as a management problem, not just an individual productivity upgrade. The argument is that managers need hands-on familiarity with AI tools, clearer goals, tighter coordination, and more explicit decisions about spend and performance expectations.
-
Tweet
Mar 13, 2026
Build your own software factory
The post frames AI-enabled delivery as an organizational capability, not just a tool choice: teams can build a reusable software factory around cloud agents and shared workflows. Useful signal for companies treating agentic engineering as platform strategy.
-
Tweet
Mar 13, 2026
Top performers in the agent era
The post suggests AI-native leverage will favor both ends of the spectrum: deep experts who can direct many agents in a narrow domain, and broad generalists who can orchestrate cross-functional execution. Helpful framing for hiring and org design as agent usage scales.
-
Tweet
Mar 13, 2026
Agents will drive software procurement
This is a useful reframing of go-to-market for AI-era products: agents become a procurement and implementation layer, not just end users. If a product cannot be provisioned and operated cleanly through APIs, it risks being excluded from agent-driven workflows regardless of brand preference.
-
Tweet
Mar 12, 2026
Takeaways on Long-running Autonomous Coding Agents
A compact synthesis of how teams are making coding agents reliable over longer time horizons. The throughline is that agent performance depends less on chat continuity and more on organizational structure, written artifacts, and objective validation.
-
Tweet
Mar 12, 2026
Planning becomes the bottleneck when building gets cheap
The post argues that cheaper implementation changes the operating model: for many product questions, a quick prototype can now be less costly than a long planning discussion. That shifts the bottleneck toward judgment, experiment design, and deciding what should be validated empirically.
-
Tweet
Mar 12, 2026
There’s never been a better time to be a PM
The thread frames AI as leverage for product managers across several adjacent tasks, not just drafting documents. The broader point is that cheaper access to context, prototypes, and implementation can let PMs compress feedback loops and contribute more directly across product, engineering, and go-to-market work.
-
Tweet
Mar 12, 2026
AI shifts productivity from people to processes
The post frames AI adoption around process compression rather than headcount replacement. The core idea is that the larger gain comes from reducing the waiting time between departments and systems, not just making one person work faster.
-
Tweet
Mar 11, 2026
The New Founder Mode
The article frames founder mode as a response to organizational drift rather than a generic call for founder intensity. Its practical point is that leaders need tighter loops with the work when scale starts to add distance, handoffs, and slower decision-making.
-
Tweet
Mar 11, 2026
Anti-fragile Infrastructure
A useful framing for AI coding adoption that shifts the discussion from model capability to deployment architecture. The main idea is that if agents are going to write and ship code, the surrounding platform needs to make testing, rollback, staged release, and isolation routine rather than exceptional.
-
Tweet
Mar 11, 2026
Productive Individuals Don't Make Productive Firms
The post frames AI adoption as an organizational design problem rather than a simple tool rollout. Its core claim is that durable value comes from rebuilding coordination, incentives, and decision processes around AI, not just making individual contributors faster.
-
Tweet
Mar 11, 2026
You’ll be able to fork agentic orgs
A compact way to describe what changes when work gets encoded into agentic systems instead of staying embedded in informal organizational process. The post suggests that once operating patterns become inspectable and executable, they may also become easier to copy, adapt, and improve.
-
Tweet
Mar 11, 2026
The new IDE is a tool for compound engineering
This extends the bigger-IDE idea into a product requirement: teams need an interface for orchestrating many agents, mixing models and harnesses, and attaching them to a queue of work. The post is useful because it frames the opportunity less as a single model breakthrough and more as an operations and UX problem for compound engineering.
-
Tweet
Mar 11, 2026
We’re going to need a bigger IDE
Useful reframing for AI coding tools: the IDE does not disappear, it expands. The post argues that as developers move up a level of abstraction, the main unit of work shifts from individual files to agents that still need visibility, orchestration, and programming interfaces.
-
Tweet
Mar 11, 2026
Mandatory AI coding adoption can become a metrics trap
The post’s main point is about incentives, not just outages: once AI adoption becomes a tracked corporate goal, teams can end up defending the metric while quietly adding process to offset operational risk. It is a useful example of how rollout pressure, autonomy, and production guardrails interact.
-
Blog post
Mar 11, 2026
Right-Sizing Engineering Teams for AI
The piece argues that AI changes the staffing equation by amplifying output without removing the need for experienced reviewers. Its practical takeaway is that smaller teams can work well, but only if they preserve enough senior judgment relative to the amount of code being produced.
-
Tweet
Mar 10, 2026
Knowing what to ship matters more than just shipping
A concise statement of a broader organizational shift: if AI lowers the cost of implementation, the differentiator moves upstream to intent and selection. The point is not shipping more by default, but choosing better work with more discipline.
-
Tweet
Mar 10, 2026
AI adoption metrics can distract from product judgment
A useful counterweight to AI adoption dashboards: output metrics can be directional, but they are not the product. The post argues that faster building only helps if teams keep quality bars, judgment, and user outcomes in view.
-
Tweet
Mar 10, 2026
When Your Life’s Work Becomes Free and Abundant
The piece is useful on two fronts: first as a candid account of what it feels like when a core technical craft becomes dramatically cheaper, and second as a hiring argument. Agarwal’s claim is that adaptability and a builder’s disposition are becoming more predictive than pedigree or years of experience.
-
Tweet
Mar 10, 2026
Do not outsource code understanding to feedback loops
The argument is that post-deployment feedback loops are not a substitute for code review or shared system understanding. It is a useful caution for teams adopting coding agents aggressively without protecting their debugging muscle.
-
Tweet
Mar 10, 2026
AI does not justify rebuilding your CRM
This is a straightforward adoption rule: lower build costs do not remove the need for buy-versus-build discipline. Useful because AI often makes custom software feel cheaper than the operational burden it creates later.
-
Tweet
Mar 10, 2026
Developer to Fleet Commander
The post frames AI adoption as a role change as much as a tooling change: some builders are pulling far ahead, and the work increasingly looks like directing systems rather than writing every step manually. Useful as a signal that AI may increase variance across teams, not just average output.
-
Tweet
Mar 10, 2026
AI coding erodes delayed gratification
The post captures a familiar local effect of AI coding tools: they make it easier to ship the next thing, which can quietly lower standards around product judgment, refactoring, and cleanup. Useful as a reminder that speed gains are not real if teams give up discipline.
-
Tweet
Mar 10, 2026
Amazon’s AI-assisted outages prompt a deep dive
This is a concrete counterpoint to the usual productivity narrative: an engineering org is treating AI-assisted changes as a reliability concern with enough operational impact to warrant a focused incident review. The key signal is not just the outages but the admission that safeguards are still immature.
-
Tweet
Mar 10, 2026
Stripe is testing billing for LLM tokens
The signal here is productization: Stripe is treating token billing as a distinct billing shape with pricing sync, usage recording, and markup support. That suggests AI-native billing is moving from custom logic toward standard platform infrastructure.
-
Blog post
Mar 10, 2026
AI should help us produce better code
A useful corrective to the idea that AI-assisted coding must trade quality for speed. The article argues that when refactors and experiments get cheaper, teams can afford stricter standards and use agents to improve both code and decision-making.
-
Tweet
Mar 10, 2026
The Harness Is The Product. The Model Never Was
The framing shifts attention away from model selection and toward the layer around the model: tools, guardrails, context handling, and workflow design. It is a useful reminder that production agent quality is usually an environment problem before it is a frontier-model problem.
-
Tweet
Mar 9, 2026
Agent harnesses versus frameworks and raw code
The framing is useful because it separates three different implementation choices that often get lumped together: writing directly against model APIs, assembling with frameworks, and adopting a fuller harness with more built-in workflow and control. Good shorthand for discussing tradeoffs in agent stack design.
-
Tweet
Mar 9, 2026
Reviewers become more valuable as builders scale
The interesting claim here is organizational, not technical: when building gets cheaper, review quality matters more. The chart frames product/design review and engineering review as the mechanisms that keep faster builders out of trouble.
-
Tweet
Mar 9, 2026
Using skills to accelerate OSS maintenance
Practical example of turning agentic coding into a repeatable workflow: encode maintenance playbooks as reusable skills and run them through CI. The useful takeaway is less about one model and more about packaging context, verification, and handoff steps so OSS work stays consistent at scale.
-
Tweet
Mar 9, 2026
The execution layer between LLMs and tools
The diagram is a compact way to describe where agent systems actually become useful in production: not at the model or the tool boundary alone, but in the execution layer that manages how the two interact. Useful framing for teams designing agent runtimes rather than single-shot prompts.
-
Tweet
Mar 8, 2026
Building for trillions of agents
Levie sketches the supporting stack he expects around large-scale agents: sandboxed execution, tool access, memory, wallets, identity, and governance. It is useful as a map of the infrastructure and workflow changes companies may need as agents move from demos into real operating environments.
-
Tweet
Mar 7, 2026
AI makes product judgment more important
The core claim is that AI reduces the cost of building, which raises the value of people who can decide what to build, how to sequence it, and how to frame it coherently. It is a useful articulation of how AI can shift status and bottlenecks inside product teams.
-
Tweet
Mar 6, 2026
5% of merged PRs now come from background agents
Concrete adoption metric: background agents now account for 5% of merged pull requests. The post also points to the surrounding workflow, including Slack as the coordination layer and local dev environments with Claude Code.
-
Blog post
Mar 5, 2026
Services: The New Software
This piece offers a practical adoption lens: as model capability rises, value shifts from selling software seats to delivering finished work. It is especially relevant for teams evaluating where outcome-based AI services can replace existing outsourced processes first.
-
Blog post
Mar 3, 2026
Harness engineering
The post argues that the leverage point is often the harness around the model: tools, instructions, verification loops, and environment design. Useful framing for teams that are focusing too narrowly on model selection alone.
-
Other
Feb 24, 2026
Stripe's annual letter shows how AI is accelerating business formation
This is a useful infrastructure signal: core business systems are being adapted for agent participation, not just human operators. That suggests more new companies will be built with AI embedded in the operating stack from the start.
-
Blog post
Feb 19, 2026
Stripe’s Minions, Part 2: blueprints, rules, and MCP at scale
This follow-up focuses on the operating model behind the agents: blueprints, scoped rule files, shared MCP infrastructure, and bounded CI loops. It is a useful example of the harness needed to make coding agents repeatable at large-company scale.
-
Blog post
Feb 13, 2026
The Final Bottleneck
The piece argues that faster code generation does not remove human responsibility; it amplifies the downstream bottlenecks around review, comprehension, and accountability. It is a useful framing for teams finding that code volume is growing faster than their ability to safely absorb it.
-
Blog post
Feb 11, 2026
The AI Vampire reframes AI as a leverage engine
Yegge frames AI less as a simple productivity tool and more as a force multiplier with side effects. The post is especially useful for thinking about leverage, exhaustion, and who captures the value of faster output.
-
Blog post
Feb 9, 2026
Minions: Stripe’s one-shot, end-to-end coding agents
Stripe says Minions already generate more than a thousand merged pull requests per week. The notable pattern is end-to-end agent implementation paired with normal human review before merge.
-
Blog post
Nov 10, 2025
10 Highlights from USV's 2025 Annual Meeting
The piece is useful as a map of how one major venture firm is framing AI across multiple markets at once. It shows AI appearing both inside firm workflows, where it helps capture and publish more ideas, and inside portfolio companies using it to reshape domains like geothermal discovery and medical triage.
-
Blog post
Mar 13, 2026
Writing code is cheap now
This piece reframes the bottleneck: generating code is now cheap, but ensuring quality, correctness, and maintainability is still expensive. It is a useful guide for teams adapting engineering habits to agent-driven workflows.
-
Tweet
Mar 20, 2026
Threading OpenAI agents
This entry captures an early operator view of OpenAI agents and the workflows they may unlock. The main theme is that agent capability changes how people delegate, supervise, and structure software work.
A useful organizational tactic for applied AI teams: shorten the distance between technical builders and domain experts until workflow discovery becomes a daily habit. The underlying point is that many valuable automation opportunities only surface through constant, low-friction collaboration.