Claude Launches Its Agent Platform, ChatGPT Gets a Filing Cabinet, and Copilot Rewires Office

#3 Perplexity Computer goes enterprise with 20 models and Slack connector

Perplexity launched Computer for Enterprise at its Ask 2026 conference. The platform routes tasks across 20 AI models including Grok, creates subagents for specialized steps, and connects natively to Slack and Snowflake. Max tier ($200/month) required. Perplexity's ARR hit $450 million in March - a 50% jump driven by this pivot to agentic products.

Executive angle: If your team is already on Perplexity Max, test the Slack connector on one repeatable research workflow before expanding further.

Read more ->

#4 Google Gemma 4: enterprise-grade open models under Apache 2.0

Google released the Gemma 4 family - open-weight models up to 31B parameters - available on AI Studio, the Gemini API, and Google Cloud. Native function-calling, structured JSON output, and agentic workflow support. Apache 2.0 license means commercial deployment with no royalty obligations.

Executive angle: If your team runs internal chatbots or data classification at volume, self-hosting Gemma 4 eliminates per-token API costs entirely - worth a cost comparison this quarter.

Read more ->

#5 Perplexity faces 135-page class action over user data sent to Meta and Google

A federal lawsuit filed in San Francisco alleges Perplexity embedded Meta Pixel, Google Ads, DoubleClick, and Meta Conversions API trackers in its platform code, allegedly transmitting user conversations including prompts, email addresses, and device identifiers to Meta and Google for ad targeting. The lawsuit claims this occurred even in incognito mode.

Executive angle: If your team shares sensitive client or proprietary data with Perplexity, switch to Claude or a self-hosted model until Perplexity makes a formal public disclosure.

Read more ->

02 / Top AI Updates | 3 Deep Dives

# 01 / 03

Claude Managed Agents: how to build and deploy production AI agents without the infrastructure headache

On April 8, Anthropic launched Claude Managed Agents in public beta - a hosted platform for building and running autonomous AI agents powered by Claude, without having to build your own backend infrastructure. The platform handles sandboxed code execution, long-running sessions, agent checkpointing, credential management, scoped permissions, multi-agent coordination, and full execution tracing. Three major SaaS companies confirmed 10x improvements in time-to-ship at launch: Notion, Rakuten, and Asana.

Business Impact

Eliminates: The need to build and maintain custom backend infrastructure to run production AI agents

Time back: Weeks to months of engineering time - Anthropic says prototype to production in days, not months

Best for: Executive teams running repeatable internal workflows like report generation, lead routing, or content moderation

Executive Action Steps

• Go to console.anthropic.com and request Claude Platform access if you do not have it already.
• Read the Claude Managed Agents quickstart - all endpoints require the managed-agents-2026-04-01 beta header.
• Pick one repeatable internal workflow to automate first (weekly report generation, lead routing, content moderation) and set your session time limits and scoped permissions before any agent goes live.
• Start with Console (no-code) to validate the workflow before moving to CLI or API - pricing is standard Claude token rates plus $0.08 per active session-hour.

Why it matters: Managed Agents removes the last barrier to production AI agents for executive teams - the infrastructure build. With Notion, Rakuten, and Asana already reporting 10x faster time-to-ship, the platform is proven at scale. The Console no-code path means you do not need an engineering team to validate your first agent workflow.

Deep dive: Claude Managed Agents ->

# 02 / 03

ChatGPT's April overhaul: File Library, GPT-4o retirement, and a cleaner model lineup

OpenAI shipped three simultaneous changes in the first week of April that reshape how ChatGPT works as a daily work tool. GPT-4o is fully retired. The File Library is now live for paid subscribers with permanent, searchable document storage. And the new model lineup - GPT-5.3 Instant, GPT-5.4 Thinking, and GPT-5.4 Pro - gives every task a purpose-matched model.

Business Impact

Eliminates: The re-upload cycle for reference documents and the guesswork of model selection

Time back: 20-35 minutes per client session using File Library plus time lost to re-uploading documents across sessions

Best for: Executives and teams who use ChatGPT daily for drafts, analysis, research, and client work

Executive Action Steps

• Update any API integrations or custom GPTs referencing GPT-4o - migrate to GPT-5.3 Instant for speed-first tasks.
• Open ChatGPT and upload your 3 most-used reference documents to start populating your Library.
• Test Library search by typing a description of a file you uploaded to confirm it is working.
• Match your model selection to task type: Instant for drafts and summaries, Thinking for analysis and planning, Pro for high-stakes research - and check your team's custom GPT configurations for any model pinned to GPT-4o.

Why it matters: The File Library fundamentally changes ChatGPT from a stateless tool to a persistent work environment. Combined with the model retirement, this is the biggest structural change to ChatGPT's daily workflow in over a year. Teams that adapt their setups now will have a compounding advantage over those who do not.

Deep dive: ChatGPT April Overhaul ->

# 03 / 03

Microsoft pushed Wave 3 of Microsoft 365 Copilot this week, centered on Work IQ - a set of agentic capabilities that let Copilot take multi-step actions inside Office apps without requiring step-by-step prompting. In Excel, Copilot reads across multiple named tables and generates formatted status reports with week-over-week comparisons. In Word, it restructures documents based on stated audience intent. In Outlook, it triages inboxes, drafts replies for low-priority messages, and schedules follow-ups automatically.

Business Impact

Eliminates: Manual weekly report-building in Excel and step-by-step prompting for multi-step Office tasks

Time back: 45-60 minutes of manual report-building per week in Excel alone, plus inbox management time in Outlook

Best for: Executives and teams already running Microsoft 365 who want autonomous actions inside the Office apps they use daily

Executive Action Steps

• In Excel, name your key data tables (Sales_Pipeline, Tasks_Open, Budget_Actuals) and test Copilot's cross-table summary prompt.
• In Word, open a long internal document and prompt Copilot: 'Rewrite the executive summary for a non-technical audience.'
• In Outlook, enable Copilot Inbox Triage in Settings, then Copilot, then Priority Inbox.
• Ask your IT admin to confirm your M365 Copilot Wave 3 rollout status - not all tenants have received it yet - and assess the E7 Frontier Suite math if you currently run E5 plus separate Copilot and Entra licenses.

Why it matters: Wave 3 moves Copilot from a prompt-response assistant to an autonomous agent inside your existing Office environment. The Excel cross-table reporting alone replaces a weekly manual process most executive teams still run by hand. The Outlook triage capability is the closest thing to a real AI chief of staff that Microsoft has shipped.

Deep dive: Copilot Wave 3 ->

03 / Pro Tip

Build a persistent client brief system using ChatGPT File Library

Tools: ChatGPT File Library | Time: 30 min | Level: Beginner

The single most frustrating thing about using ChatGPT for client work has been re-uploading the same reference documents at the start of every session. The File Library ends that. Upload your standard client brief template, the last 3 deliverables you created for that client, and your company's tone and style guide. All three save to Library automatically. Setup time: 30 minutes. Estimated time saved per client session: 20-35 minutes.

The Prompt

Why it works: The File Library transforms ChatGPT from a stateless tool into a persistent client work environment. By anchoring every session to the same reference documents, you get consistent voice, faster output, and fewer revision cycles - without any re-uploading.

Best for: Executives and account managers who use ChatGPT regularly for client deliverables, proposals, and communications

Deep dive: Build a persistent client brief system using ChatGPT File Library ->

04 / Productivity Gem

Copilot in Excel: turn named tables into an automated weekly status report

Tools: Copilot in Excel | Saved: 2 hrs/week

Microsoft Copilot Wave 3 in Excel can now read across multiple named tables in a single workbook and generate a formatted status report with delta comparisons from the prior week. The setup takes 15 minutes once; after that, the report runs in under 5 minutes every week.

• Name your tables inside Excel: Insert, then Table, check 'My table has headers', then right-click, then Table, then Table Name. Use clear names like Sales_Pipeline, Tasks_Open, and Budget_Actuals.
• Open Copilot in Excel and run the reusable prompt below. Repeat each Monday morning for a fully formatted status report with week-over-week comparisons - no manual assembly required.

The Prompt

Time savings: Saves an estimated 45-60 minutes of manual report-building per week

Deep dive: Copilot in Excel: turn named tables into an automated weekly status report ->

05 / Health Tip

Build a Living Research Report with Perplexity (and update it every week)

Tools: Perplexity

This project takes 30-45 minutes the first session and 15 minutes each week after that. Kids pick a topic they genuinely care about - a sports team's season stats, how bees communicate, the history of video game controllers - and use Perplexity (free, no account required) to build a structured research report across 5 key questions. What makes it a Living Report is the update cycle: every week, they add one new section based on a new question they did not know to ask before. By week 4, they have a 4-section research report they built themselves.

Try Asking

Core lesson: The real research skill is knowing what question to ask next - not just finding answers to questions you already have.

Deep dive ->

That is the April 10 issue. Three platforms made meaningful moves this week - and two of them (Claude Managed Agents, Copilot Wave 3) are directly actionable for your workflows today, not someday. The Perplexity lawsuit is worth watching: any tool that markets incognito mode and then transmits prompts to ad networks has a trust problem that goes beyond legal exposure. Use Claude or a local model for anything sensitive until Perplexity clears that story.

Executives who act on this issue this week will have a persistent client brief system in ChatGPT, a working Excel status report that runs itself on Mondays, and a clear decision on whether Claude Managed Agents is the right platform for their first production agent workflow.

Forward this to one executive you respect. They will thank you next week.

Premium Upgrade