Managerbot, Glasswing, and the Proactive Agent Era

The Hook

The agentic shift is no longer theoretical — it is running on your merchant’s Square dashboard. Block launched Managerbot this week, a proactive AI agent embedded in Square that monitors inventory, schedules staff, and drafts marketing campaigns before sellers think to ask. Simultaneously, Anthropic unveiled Project Glasswing and Claude Mythos Preview — a restricted frontier model that autonomously identified thousands of zero-day vulnerabilities, including a 27-year-old flaw in OpenBSD and a 16-year-old bug in FFmpeg that automated scanners had missed across 5 million test runs. The same week, Accenture invested in Replit and Belitsoft’s market forecast landed: Gartner projects 40% of enterprise applications will include task-specific AI agents by end of 2026, up from near zero 18 months ago. The agents are not coming. They are here, and they are proactive.

This Week’s Signal

From Reactive to Proactive: The Managerbot Shift Redefines What an Agent Is

Block’s Managerbot, now rolling out to Square sellers, is the clearest commercial signal yet that the definition of an AI agent has permanently changed. Square’s earlier AI assistant was a chatbot — it answered questions sellers asked. Managerbot is fundamentally different. It watches business data continuously, surfaces problems before sellers detect them, and proposes specific actions: reorder this SKU before the weekend rush, adjust Thursday night’s shift, launch a re-engagement campaign targeting customers who haven’t returned in 45 days.

According to Block’s head of product Willem Ave, the core shift is from a question box to a task queue. “You assign tasks to Managerbot, and that could be based on data, an insight, or a signal from your business.” The agent acts first. The seller approves or adjusts. The workflow is inverted.

This is not a luxury enterprise feature. It is shipping to millions of small businesses — restaurants, retailers, service providers — through an existing subscription they already pay for. The democratization of proactive agentic behavior is arriving through commerce infrastructure, not developer tooling. The operator implication is direct: if a payments platform can build this, the same proactive loop applies to any domain where a business generates continuous structured data and acts on it repeatedly.

Alongside Managerbot, Accenture announced a strategic investment in Replit on April 9 to accelerate enterprise adoption of AI-driven software development. The combination — proactive operational agents on the SMB side, AI-accelerated software creation on the enterprise side — frames the week’s signal: agents are now a default product feature, not an advanced capability.

3 Operator Playbooks

1. Map Your Business’s Continuous Signal Loop and Build the Agent on Top of It

Managerbot works because Square already holds the data: sales by SKU, shift logs, customer visit frequency, ticket times. The agent adds a continuous monitoring layer and a decision-proposal interface. Most businesses already have equivalent data sitting in disconnected tools — POS, CRM, HR scheduling software, email marketing dashboards.

Your move: Identify the one workflow in your operation where a human currently checks a dashboard, forms a judgment, and initiates an action on a recurring cadence. That is your Managerbot candidate. Map the data inputs, define the three most common action types, and scope an agent that monitors, proposes, and logs approvals. The approval gate is not a concession to caution — it is how you build trust in the loop before removing friction later.

2. Use Glasswing’s Controlled-Release Model as Your Agentic Rollout Template

Anthropic’s Project Glasswing did not release Claude Mythos Preview publicly. It enrolled 12 named launch partners — AWS, Apple, Cisco, CrowdStrike, Google, JPMorganChase, the Linux Foundation, Microsoft, NVIDIA, Palo Alto Networks, Broadcom, and Anthropic itself — plus 40 additional critical infrastructure organizations, with Anthropic committing up to $100 million in usage credits and $4 million in direct donations to open-source security organizations. The model is too capable to release without a structured trust-building phase.

That structure is a deployable pattern. When you ship a high-capability agent into a domain where errors are costly — financial decisions, compliance workflows, customer-facing actions — start with a defined cohort of internal power users, instrument every action, build a review cadence, and expand access only after the error profile is characterized.

Your move: Before your next agent launch, define the controlled release: who are your 12 launch partners? What is the review mechanism? What does the expansion trigger look like? Glasswing is a $100M cybersecurity initiative, but the rollout architecture scales down to a five-person ops team.

3. Invest in the Accenture-Replit Stack Before Your Engineering Backlog Compounds

Accenture’s April 9 investment in Replit targets a specific enterprise bottleneck: organizations that want to move fast on agentic product development but lack the internal engineering capacity to build at the required pace. The partnership combines Accenture’s enterprise scaling expertise with Replit’s AI-driven development platform, where teams generate functional code from plain-language instructions.

Belitsoft’s 2026 forecast released the same week quantifies the urgency: agentic AI has risen from 13.0% to 17.1% as a top-ranked enterprise technology priority in a single year, with the OutSystems survey of 1,900 global IT leaders showing 49% describing their agentic AI capabilities as advanced or expert — up from a small fraction the prior year.

Your move: If your team has an agent roadmap but an engineering backlog that stretches it 12 months out, evaluate whether AI-assisted development tooling (Replit, Cursor, or similar) can compress that timeline. The competitive gap between teams using AI-accelerated development and those that are not is already measurable. The Accenture-Replit partnership is an institutional signal that this toolchain is now enterprise-safe, not just developer-experimental.

Steal This

Proactive Agent Design Canvas — 5-Field Spec

Before you build a Managerbot equivalent for your domain, fill in these five fields. Each one is a failure mode if left undefined.

PROACTIVE AGENT DESIGN CANVAS

1. DATA SIGNAL
   What continuous data stream does the agent monitor?
   (Example: daily SKU velocity, customer return cadence, support ticket volume)

2. TRIGGER CONDITION
   What specific pattern causes the agent to surface a proposal?
   (Example: inventory below 7-day projected demand, ticket volume >15% above 30-day mean)

3. ACTION TYPES (max 3 to start)
   What are the three actions the agent can propose?
   (Example: draft reorder PO, adjust shift coverage, initiate discount campaign)

4. APPROVAL GATE
   What must a human confirm before the action executes?
   (Example: any action above $500, any communication to >100 customers)

5. CONFIDENCE FLOOR
   Below what evidence threshold does the agent stay silent?
   (Example: triggers only on 14+ days of data, not on single-day anomalies)

Copy this into your product spec or agent design doc before writing a line of code. Most failed agent pilots skip fields 4 and 5. That is where trust collapses.

The Bottom Line

This week’s pattern is proactive agency at commercial scale. Block shipping Managerbot to millions of small businesses, Anthropic restricting Mythos Preview to 52 trusted partners, and Accenture backing Replit’s enterprise coding platform — these are not isolated product announcements. They are three simultaneous answers to the same question: how do you deploy agents responsibly when the capability has outpaced the governance? Managerbot answers with approval gates. Glasswing answers with controlled cohorts. The Accenture-Replit partnership answers with enterprise-safe infrastructure. Gartner’s 40% of enterprise applications by end of 2026 is not a projection to plan toward — it is a baseline to measure against. The operators who ship proactive agents with real approval architecture this quarter are not ahead of the curve. They are on it.

AI Agent Insider is published by Digital Forge Studios Inc.