<p><strong>TLDR:</strong> Replit closed a $400M Series D at a $9B valuation, tripling from $3B just six months ago. Lovable crossed $400M ARR in February after adding $100M in one month with 146 employees. Codex had a stability incident today and was restored. The GPT 5.4 vs Opus 4.6 debate is heating up, with practitioners splitting on which to use and when.</p>
<hr>
<h2>Replit raises $400M at $9B valuation</h2>
<p>Six months after hitting $3B, Replit closed a $400M Series D led by Georgian Partners, with a16z, Coatue, YC, and Accenture Ventures participating. The company reported $150M ARR at the September round and is now targeting $1B ARR by year-end. CEO Amjad Masad credits the growth to a deliberate pivot away from professional developers toward non-programmers — a bet that took nine years to pay off. Agent 4 launched this week alongside the raise.</p>
<h2>Lovable crosses $400M ARR, $100M added in February alone</h2>
<p>Lovable went from $100M ARR in July 2025 to $400M by February 2026, adding the last $100M in a single month with 146 employees — roughly $2.77M ARR per head. Enterprise adoption is real: Klarna and HubSpot are customers, and the company claims over half of Fortune 500 companies are using the platform. Valued at $6.6B. The numbers are hard to ignore even if the vibe-coding category still makes some engineers uncomfortable.</p>
<h2>Codex outage, now resolved</h2>
<p>Codex went down today and was restored after a reset. No detailed postmortem yet. Worth noting given how many people are running Codex as a primary tool — the dependency risk is real when your CI pipeline runs through a third-party agent service.</p>
<h2>GPT 5.4 vs Opus 4.6: practitioners are splitting</h2>
<p>The practitioner consensus forming in the wild: GPT 5.4 fast mode wins on speed, reasoning, and complex multi-file work; Opus 4.6 wins on frontend aesthetics and UI quality. A structured 100-task benchmark gave Claude Code 78% first-pass accuracy vs Cursor’s 73%, with an 18% speed edge on complex tasks. Cost math favors Cursor for simple tasks ($0.19/task) and Claude Code for complex ones ($0.87 vs $1.14). Several engineers report dropping Opus entirely for GPT 5.4 fast, while others are staying Claude-first for frontend work. Opus 5 is the obvious wildcard.</p>
<hr>
<h2>Also notable</h2>
<ul>
<li><strong>VS Code 1.112 Insiders:</strong> Claude agent mode graduated from preview to GA; MCP servers and agent plugins are now toggleable at workspace and global levels.</li>
<li><strong>Cursor Marketplace:</strong> Over 30 new plugins added, including Atlassian, Datadog, GitLab, Glean, and PlanetScale — all bundling MCP capabilities with agent instructions.</li>
<li><strong>Gemini CLI plan mode:</strong> Now ships read-only by default, with ask_user tool for clarification, read-only MCP support, and automatic routing to higher-reasoning Pro models during planning.</li>
<li><strong>GitHub Copilot for JetBrains:</strong> Custom agents, sub-agents, and plan agent now GA; <a href="http://AGENTS.md" target="_blank" rel="noopener nofollow">AGENTS.md</a> and <a href="http://CLAUDE.md" target="_blank" rel="noopener nofollow">CLAUDE.md</a> instruction file support added; thinking panel for extended-reasoning models in public preview.</li>
<li><strong>Gemini Code Assist:</strong> Agent Auto Approve mode shipped alongside inline diffs and custom slash commands — Google’s own docs call auto-approve “extremely dangerous and never recommended” while the blog promotes it.</li>
<li><strong>Claude for Excel and PowerPoint:</strong> Cross-app context sharing now in beta for paid users; Skills feature lets teams save repeatable financial analysis workflows as one-click actions.</li>
<li><strong>Azure Skills Plugin:</strong> Microsoft packaged 19+ Azure skills, the Azure MCP Server (~200 tools), and Foundry MCP Server into a single install compatible with Copilot, Claude Code, and other agents.</li>
<li><strong>Databricks Genie Code:</strong> GA in Databricks workspaces; integrates with Unity Catalog for data lineage and governance, claims 2x improvement over a leading coding agent on real-world data science tasks.</li>
<li><strong>GitHub CLI v2.88.0:</strong> You can now request a Copilot code review directly from the terminal with <code>gh pr edit --add-reviewer @copilot</code>.</li>
<li><strong>Devin enterprise usage:</strong> Cognition reports more Devin sessions and merged PRs in the first two months of 2026 than in all of 2025.</li>
<li><strong>Sam Altman on GPT-5.5:</strong> “We are training right now — on the first site in Abilene — what I think will be the best model in the world. Hopefully by a lot.”</li>
<li><strong>Antigravity rate limits:</strong> At least one user hit a week-long usage cap after an hour-long session on AI Pro, suggesting the new limits are tighter than before.</li>
<li><strong>Crypto GitHub commits:</strong> Down 75% year-over-year with active developers cut in half — the less obvious read is that one developer with Cursor and Claude now ships what took five people in 2023.</li>
</ul>


AI coding assistants, code generation tools, and vibe coding workflows.

Agentic Digest

Replit closed a $400M Series D at a $9B valuation (up from $3B six months ago), targeting $1B ARR by year-end after pivoting to non-programmers. Lovable hit $400M ARR with 146 employees, adding $100M in February alone, with Klarna and HubSpot among enterprise customers. OpenAI's Codex suffered a brief outage. Practitioners are splitting between GPT 5.4 fast mode (speed, complex reasoning) and Claude Opus 4.6 (frontend/UI quality), with a 100-task benchmark giving Claude Code 78% first-pass accuracy vs Cursor's 73%. Other notable updates: VS Code 1.112 graduates Claude agent mode to GA, Cursor Marketplace adds 30+ plugins, GitHub Copilot for JetBrains ships custom agents, Gemini Code Assist adds Auto Approve mode, and Devin reports more sessions in the first two months of 2026 than all of 2025.

Replit hits $9B, Lovable adds $100M in a single month