AcceleratingBiz®

Growth CEO’s AI Briefing · Edition #3

Agents Move in

Thursday, April 9, 2026 · 3 Developments · 1 Perspective · 1 Case Study · 1 Resource

Designed by AcceleratingBiz®. Curated, synthesized, and produced by Claude agents. AI-assisted production means occasional errors may appear.

The theme across today's content is consistent: AI agents are moving from concept to operational infrastructure. Meta shipped a new model designed for distribution over benchmarks, Anthropic opened a managed platform so any developer can deploy agents in days, and Perplexity's revenue surged 50% in a single month after pivoting to agent products. Meanwhile, Every published two complementary pieces showing what it actually looks like when agents become half your workforce.

  • Meta's comeback play. Muse Spark, the first model from Meta Superintelligence Labs, is competitive on reasoning but trails on coding; the real story is that it ships inside apps used by 3 billion people daily.
  • Anthropic's agent infrastructure. Claude Managed Agents entered public beta, letting developers define tasks and guardrails while Anthropic handles everything else for $0.08 per session hour.
  • Perplexity's agent-driven revenue spike. ARR topped $450 million in March after the company shifted from AI search to autonomous agents and usage-based pricing.
  • The agent-native infrastructure thesis. TheSequence argues that every layer of software infrastructure needs to be reimagined for agents as first-class users.
  • Every's agent org chart. The media company now runs four custom Notion agents handling prioritization, meeting-to-task conversion, OKR planning, and growth scorecards across its 25-person team.
  • Every's agent etiquette lessons. Building on their earlier coverage, Every explored the practical tensions of agent-human collaboration, including when to message the agent versus the human, and how managing agents forces everyone to become a manager.
  • Claude Cowork exits preview. Anthropic moved Cowork from research preview to general availability, adding enterprise features including role-based access controls and usage analytics.

The pattern worth noting: the companies gaining traction aren't just building better models. They're building the infrastructure that makes agents deployable, manageable, and billable.

Developments

Competition

April 8: Meta Launches Muse Spark, Its First Model from Superintelligence Labs

On April 8, Meta debuted Muse Spark, the first model from Meta Superintelligence Labs, the AI division assembled nine months ago after Zuckerberg acquired Scale AI for $14.3 billion and installed Alexandr Wang as chief AI officer. The multimodal model handles voice, text, and image inputs, with a multi-agent "Contemplating" mode for complex reasoning. It scored a 53 on the Artificial Analysis Intelligence Index, debuting at #4 behind Gemini 3.1 Pro, GPT-5.4, and Claude Opus 4.6. Notably, Muse Spark is proprietary, marking a significant departure from Meta's open-source Llama strategy.

So what

Meta doesn't need the best model if it has the best distribution. Muse Spark ships inside apps used by 3 billion+ people daily, which is a fundamentally different competitive strategy than selling API access. For growth companies evaluating their AI stack, the question is whether Meta's distribution advantage eventually commoditizes standalone AI products. Watch whether Meta opens an API for third-party developers; if it does, it could offer a lower-cost alternative to OpenAI and Anthropic for applications where "good enough" reasoning suffices.

Hype check

Meta acknowledges meaningful gaps in coding (trailing Claude Opus 4.6 by ~7 points on SWE-bench) and agentic workflows (15% behind GPT-5.4 Pro), yet coverage framed this as a "comeback." Competitive on multimodal perception is not competitive across the board.

Confidence: Established fact — model launch confirmed by Meta, benchmarks independently verified by Artificial Analysis, covered by Bloomberg, CNBC, Axios, and 8+ newsletter sources.

Go deeper Meta AI Blog · Axios · The Neuron

Technology

April 8: Anthropic Launches Claude Managed Agents in Public Beta

On April 8, Anthropic opened its Claude Managed Agents platform in public beta, providing a hosted infrastructure for building and deploying AI agents at production scale. Developers define the agent's task, tools, and guardrails; Anthropic handles sandboxing, state management, authentication, and orchestration. Agents can run solo for hours without losing state, with a multi-agent coordination mode also in preview. Notion, Rakuten, Asana, and Sentry are early adopters, with Rakuten reportedly deploying agents across five departments in about a week each.

So what

This is the clearest signal yet that the AI value chain is shifting from "model access" to "agent infrastructure as a service." At $0.08 per session hour plus token costs, the pricing is consumption-based, which aligns with how growth companies want to buy. If you're evaluating whether to build agent infrastructure in-house or buy it, Anthropic just made the "buy" option materially cheaper. Growth company CTOs should test this against their current agent prototyping stack this month.

Confidence: Established fact — public beta confirmed by Anthropic, pricing and early customers verified across multiple independent sources.

Go deeper Anthropic Blog · SiliconANGLE · The Rundown AI

Business

Perplexity ARR Tops $450M After Pivot from AI Search to Agents

Perplexity's annual recurring revenue surged to more than $450 million as of March 2026, a 50% jump in a single month, the Financial Times reported on April 8. The acceleration was driven by "Computer," an autonomous agent product launched in February, and a shift to usage-based pricing. The company now has over 100 million monthly active users and tens of thousands of enterprise clients. For context, Perplexity's ARR stood at $16 million just two years ago and $305 million before the pricing change.

So what

Two things are worth separating. First, the product signal: agents that execute tasks (not just answer questions) generate enough user value to sustain usage-based pricing, which is the monetization model most growth companies are experimenting with. Second, the revenue pace: $16M to $450M ARR in two years is fast, but still modest compared to Cursor ($2B) or Anthropic ($19B). Growth-stage leaders evaluating AI pricing models should study Perplexity's credit-based consumption approach as a template.

Hype check

The FT figures are company-provided estimates, not audited revenue. Usage-based pricing can inflate ARR snapshots if usage spikes are unsustained. Watch retention over the next two quarters before treating this as a durable trajectory.

Confidence: Informed perspective — ARR figure reported by the Financial Times based on company data; not independently audited.

Go deeper Financial Times · The Neuron

Perspectives

People & OrganizationDeveloping

Every Is "Half Agent Now": What Changes When AI Coworkers Need Managing

Building on their agent deployment covered in Edition #2, Every published an AI & I podcast episode exploring the operational and cultural tensions of running a company where half the workforce is AI. CEO Dan Shipper and COO Brandon Gell discussed practical questions: when should a teammate message the human versus the agent? How do you build trust in an agent's output? Their observation that "everyone is a manager now" captures a shift most companies haven't planned for. Notably, 70% of Every's staff already refer to their AI agents using gendered pronouns.

So what

The "everyone is a manager now" framing is the insight worth carrying into leadership conversations. As agents become operational, every employee needs delegation, evaluation, and coaching skills that were previously reserved for people managers. Growth company CHROs and CEOs should start thinking about agent management as a core competency, not a nice-to-have skill. Every's parallel org chart for agents is a practical model to adapt.

Confidence: Informed perspective — first-party operational account from a 25-person company actively running agents in production.

Go deeper Every (Context Window)

Case Studies

Operations & FinanceDeveloping

Every Runs a 25-Person Company on Four Custom AI Agents

Building on the agent deployment covered in Edition #2, Every published a detailed breakdown of four custom Notion agents running their operations. Anton handles daily task prioritization across the company, replacing the COO's manual routing. Max converts meeting transcripts into action items and creates tasks in Notion via Slack. A strategy interviewer agent compressed OKR planning from weeks to two days. A campaign reporter posts daily growth scorecards by pulling data from PostHog and Stripe via Notion Workers.

So what

The transferable pattern here isn't the specific agents; it's the architecture. All four agents query the same interconnected Notion databases, which means each new agent is cheaper to build than the last. Growth company COOs and ops leaders should identify three to four recurring coordination bottlenecks and prototype agents against them. Every's approach of starting simple and building on existing databases rather than installing templates is the right one.

Confidence: Informed perspective — first-party case study from Every, published in partnership with Notion but with independent editorial control.

Go deeper Every (Source Code)

Resources

Tools

April 9: Claude Cowork Exits Research Preview, Adds Enterprise Features

On April 9, Anthropic moved Claude Cowork from research preview to general availability for all paid subscribers on macOS and Windows. The update adds six enterprise features: role-based access controls, group spend limits, usage analytics, expanded OpenTelemetry support, a Zoom MCP connector, and per-tool connector controls. Cowork lets Claude manage workflows on the desktop, working with local files and applications to assemble deliverables.

So what

The enterprise features signal that Cowork is ready for team-level deployment, not just individual experimentation. Role-based access and spend limits are the specific controls IT and finance leaders need before approving broader rollouts. If your team has been piloting Cowork individually, this is the moment to evaluate a team-wide deployment. The Zoom MCP connector is worth testing for meeting-to-action workflows similar to what Every built with Notion agents.

Confidence: Established fact — announced by Anthropic, confirmed by 9to5Mac.

Go deeper 9to5Mac · Anthropic

Sources in this edition: The Neuron, The Rundown AI, Superhuman AI, The Deep View, TLDR AI, AINews, Daily AI, TheSequence, Every, TLDR Founders, TLDR Marketing, TLDR Crypto, The Signal, 9to5Mac

← Latest Editions and Search