1. [AINews] The Biggest Claude Launch of All Time — AINews

  • Why read: Provides early signal on why the "Claude Cowork" and "Computer Use" launches are seeing unprecedented adoption compared to previous LLM releases.
  • Summary: Following the Vercept acquisition, Anthropic’s launch of Claude Cowork Dispatch and expanded Computer Use capabilities has triggered the company's largest reception to date. The analysis suggests a major trend shift from simple chat interfaces to "coworking" agents that operate directly on user machines. This launch marks a transition where LLMs are no longer just answering questions but are actively performing tasks within a user's environment. For operators, this signals that the "agentic" workflow has moved from experimental to the primary mode of user interaction.
  • Link: https://www.latent.space/p/ainews-the-biggest-claude-launch

2. Today, Sierra is releasing Ghostwriter, our agent for building agents — Bret Taylor

  • Why read: Explores the next layer of abstraction in software: using conversational AI to orchestrate and build the very agents that will run customer experience.
  • Summary: Ghostwriter allows enterprises to build complex agents that chat, handle phone calls, and interact with systems of record through simple conversation rather than forms or menus. Taylor argues that just as Codex and Claude Code transformed software engineering, "agentic UIs" will transform all software platforms. The vision is a future where every enterprise app replaces its traditional web UI with an agent that performs work on the human's behalf. This moves software development toward a model of orchestration and review rather than manual configuration.
  • Link: https://sierra.ai/blog/agents-as-a-service

3. Meet the Agents at USV: Arthur, Ellie, Sally, and Friends — USV

  • Why read: A practical case study on how a top-tier venture firm is building and naming internal agents to automate institutional memory and context.
  • Summary: USV shares four key learnings from three months of building internal agents like "Arthur" to capture meeting context and discussions. They advocate for "treating agents like employees" and ensuring they live where the team already communicates (Slack/Email) rather than in separate dashboards. The post emphasizes solving one small, high-frequency problem—like meeting notes—before expanding to complex workflows. This "Building 20" approach allows teams to shape their own software tools organically as they work.
  • Link: https://twitter.com/usv/status/2036903978555105451

4. Notes from the SaaS Funeral — Reid Hoffman

  • Why read: A strategic defense of the software business model against the "AI is killing SaaS" narrative.
  • Summary: Hoffman argues that while AI reduces the "defensibility of sheer engineering labor," it does not eliminate the need for software as a living system requiring maintenance, compliance, and security. The moat for SaaS is shifting from "lines of code" to how well a system's agents are tuned to specific domain categories and proprietary data. He predicts a transition from cloud-native to AI-native models where prepay token budgets may replace traditional seat-based pricing. Ultimately, as the cost of building software drops, Jevons' Paradox suggests total demand for software will actually expand.
  • Link: https://twitter.com/reidhoffman/status/2036826631206326339

5. Thoma Bravo LP Meeting Slides: The Software Paradox — The Icahnist

  • Why read: High-level financial data showing a massive disconnect between public market "AI fear" and actual software business performance.
  • Summary: Thoma Bravo notes that software is trading at a discount to the S&P 500 for only the second time since 2008, despite fundamentals (17% growth vs 6% for the S&P) remaining strong. They identify a $3T incremental TAM as autonomous AI systems convert global labor spend into new software revenue. The firm distinguishes between "exposed" software (simple workflows, generalist knowledge) and "protected" software (deep compliance, proprietary data). For incumbents, AI is viewed as an "upside" that strengthens the value proposition rather than a pure replacement threat.
  • Link: https://twitter.com/TheIcahnist/status/2036902492080837015

6. Guillermo Rauch is "terminally online," anti-1:1s, and just getting started — Ali Rohde

  • Why read: Insights into the "founder-operator" mode of the Vercel CEO, focusing on extreme responsiveness and information dissemination.
  • Summary: Rauch advocates for being "terminally online" to stay close to customer feedback and product quality, even as a unicorn CEO. He is notably "anti-1:1" for information sharing, preferring long-form writing and larger group dissemination to maximize efficiency. By hiring high-agency "former founders," he creates space to do "unstructured work" like personal customer support and product testing. This leadership style prioritizes high-bandwidth external signal over internal bureaucratic synchronization.
  • Link: https://twitter.com/RohdeAli/status/2036870771503751621

7. AI's Bundling Moment — Tomasz Tunguz

  • Why read: Analyzes why the SaaS "unbundling" playbook (owning one workflow) is being replaced by AI "rebundling" (owning the platform).
  • Summary: Tunguz argues that the rapid 42-day cadence of model updates makes it impossible for buyers to assemble "best-of-breed" point solution stacks. Instead, buyers are seeking trusted platforms like Harvey (Legal/Professional Services) or Glean (Work AI) that can span multiple verticals. AI systems gain value by seeing how entire teams operate across different workflows, rewarding breadth over specialization. The new AI playbook focuses on selling "trust" and platform-wide integration rather than specific, isolated features.
  • Link: https://www.tomtunguz.com/2026-03-24-saas-unbundled-ai-rebundled/

8. seems obvious but: What is and isn't changing — rahul

  • Why read: A sharp framework for identifying which AI startups are "NGMI" (Not Going to Make It) based on what variables are commoditizing.
  • Summary: The author lists four rapidly changing variables (context windows, reasoning, benchmarks, cost per token) and four static variables (human behavior, integrations, infrastructure, CPU performance). Startups focusing on context optimization or custom multi-agent harnesses are "NGMI" because the underlying models will eventually subsume those "hacks." Conversely, "WAGMI" (We're All Gonna Make It) companies focus on product/UI, customer acquisition, and speed of agent verification loops. The winning strategy is meeting users where they are rather than inventing new "context graphs."
  • Link: https://twitter.com/rahulgs/status/2036857870042411438

9. How to use AI for your next job interview — Lenny Rachitsky

  • Why read: A highly practical deployment of a "Claude Code–based coach" for professional development.
  • Summary: Lenny's team researched 30 job seekers to build a plug-and-play AI interview coach. The tool goes beyond resume polishing to offer mock interviews that "push back," company-specific question prediction, and salary negotiation scripts. It effectively "mines" a user's experience for stories they didn't know they had by analyzing past work. This represents a shift in AI utility from "writing assistant" to "behavioral coach" and strategic negotiator.
  • Link: https://twitter.com/lennysan/status/2026369257983103210

10. 2026 State of GTM | Benchmarks & Data — OnlyCFO's Newsletter

  • Why read: Critical data showing an acceleration in revenue growth for <$100M ARR companies driven by AI-natives.
  • Summary: The ICONIQ 2026 report shows that while public software stocks are struggling, smaller AI-native companies are seeing a growth "re-acceleration." However, NRR (Net Revenue Retention) remains a lagging and declining metric for "legacy" software firms, leading to valuation compression. The report notes that AI agents (e.g., Paraglide for AR) are starting to replace high-volume, repetitive manual labor in finance and operations. For CFOs, the focus has shifted entirely to "revenue durability" and whether a company is AI-native or just "legacy" with an API.
  • Link: https://onlycfo.news/p/2026-state-of-gtm-benchmarks-data

Themes from yesterday

  • The Great Rebundling: Buyers are moving away from point solutions toward integrated "Work AI" platforms as model velocity makes custom stacks unmanageable.
  • Agentic Infrastructure: "Building agents that build agents" and internal team agents (named and treated like employees) have moved from concept to standard operating procedure.
  • The SaaS Moat Debate: Consensus is forming that SaaS isn't dead, but its moat is shifting from "lines of code" to domain-specific agent tuning and "terminally online" customer responsiveness.
  • Economic Bifurcation: A clear split is emerging between AI-native companies with accelerating growth and legacy software companies facing NRR decline and valuation compression.