One-Line Summary

Google pushes Search toward agentic AI and lightweight video creation, while DeepSeek’s price cut and fresh security lessons reset budgets and guardrails for teams.

Big Tech

Google recasts Search as an AI agent

Google is changing how Search works by merging AI Overviews and AI Mode into a more conversational experience and introducing an intelligent search box for longer, more complex questions; Google also says Gemini 3.5 Flash becomes the engine behind AI Mode. CNET reports Google frames this as the biggest search-box upgrade in over two decades and highlights rising AI Mode usage. ¹

The new interface supports uploads like photos and PDFs, can reference open Chrome tabs, and shifts seamlessly from an AI Overview to a chat-style follow‑up in AI Mode. Google also shows dynamic, interactive widgets and a move toward “agentic” help — monitoring topics and suggesting bookings with links to finalize — with capabilities rolling out starting this summer, across 200 markets and 98 languages, according to its briefing to reporters. ¹

Amid this, The New Stack contrasts Google’s new Spark personal agent with the surging interest in OpenClaw, outlining how agent platforms and the Model Context Protocol (MCP) are shaping how assistants connect to tools and data. The takeaway: competition is shifting from one-off chatbots to end‑to‑end agents that can carry out tasks. ²

Separately, Qoo Media reports Google’s Gemini Omni Flash now generates short videos directly from text, images, audio and clips, and offers text‑based editing inside the Gemini app, YouTube Shorts and Google’s Flow creative platform. The initial limit is 10‑second clips, with identity‑verified avatars and SynthID watermarking applied to outputs. ³

Industry & Biz

DeepSeek makes 75% price cut permanent on V4‑Pro

DeepSeek says it will keep a 75% discount on its flagship V4‑Pro model, holding prices for developers at a quarter of prior levels rather than letting the promotion expire at the end of May. Bloomberg notes the move is set to intensify competition with global peers. ⁴

For teams watching token bills, a stable discount alters vendor comparisons and could expand access for pilots and internal tools, especially where latency and context caching are acceptable trade‑offs. The price change is official via DeepSeek’s pricing page cited by Bloomberg. ⁴

AI security gaps surface, with Google billing incidents cited

TechCrunch highlights guidance from Google Cloud COO Francis de Souza: treat AI security as a platform issue from day one, avoid “shadow AI,” and pair any AI strategy with strong data and security strategies. He describes a faster threat window and argues for AI‑native, agentic defense overseen by humans. ⁵

The piece also cites reporting that some developers were hit with five‑figure charges after compromised API keys accessed Gemini services, and that revoking older Google API keys can take up to 23 minutes to propagate — versus roughly five seconds for service‑account credentials and about a minute for AQ‑prefixed keys — creating a risk window for data exfiltration. Google refunded affected users in cited cases, TechCrunch notes. ⁵

New Tools

Gemini Omni Flash: 10-second AI video with safeguards

Qoo Media reports that Gemini Omni Flash creates up to 10‑second videos from text, images, audio and short clips, and supports simple text‑driven edits. It’s accessible via the Gemini app, YouTube Shorts and Google’s Flow. ³

Outputs carry SynthID watermarking, and creating personal avatars requires identity verification, which aims to reduce deepfake misuse. As with other generators, fidelity depends on prompt detail. ³

Claude CFO Skill Stack: closed-loop reforecasting in Claude

Forbes spotlights a “Closed‑Loop CFO Skill Stack” that runs inside Claude Projects, adding a CFO Brain, memory layer and audit trail to speed monthly or quarterly reforecasts. Users upload financial datasets and prompt Claude to update forecasts and document reasoning. ⁶

For small finance teams, this turns repetitive rebuilds into repeatable runs, with traceability for decisions — though it still requires careful data preparation and prompt design. ⁶

Community Pulse

Hacker News (188↑) — Commenters split between seeing real savings from the discount (aided by context caching) and skepticism that company spending habits reflect careful cost–benefit analysis. ⁷

"Anecdotally it's costing the same or less than the typical coding subscription, due to the discount and the power of context caching." — Hacker News ⁷

"I'm doubtful that the companies telling their employees to burn more tokens are doing careful evaluations of cost versus benefit. People on an expense account don't shop around much. Maybe they'll penny-pinch later after running through their AI budgets?" — Hacker News ⁷

What This Means for You

Google’s shift to a conversational, agent‑style Search means research, planning and customer support workflows can live inside Search with follow‑up chat, uploads and app context — less tab‑hopping, more “get it done” in one place. If your work touches SEO, expect answer formats and entry points to keep evolving as AI Mode becomes the front door. ¹

DeepSeek’s 75% permanent discount is a nudge to revisit model mixes: run the same prompt suites you already use and compare cost per task, latency and context limits. For teams paying per token, savings on routine drafting or coding assistance can free budget for higher‑end models where quality gains matter. ⁴

Security needs an immediate tune‑up: rotate or restrict any exposed API keys, prefer service‑account or newer AQ‑prefixed keys where available, and enforce budget alerts and caps. The 23‑minute revocation window cited for older keys shows why detection-and-response at machine speed — and clear ownership — are now leadership issues, not just IT chores. ⁵

Short‑form AI video is becoming practical for social posts, product explainers and quick training clips. Try a 10‑second concept to gauge quality and brand fit before committing to longer formats or paid workflows. ³

Action Items

1. Test Google’s new AI search flow: In AI Mode (if available), run a multi‑step query, upload a PDF or image, and use follow‑up chat to see if it reduces switching between tabs.
2. Make a 10‑second sample with Gemini Omni Flash: Use the Gemini app or YouTube Shorts to generate a short clip from text and an image; compare quality to your current editing process.
3. Price‑check DeepSeek V4‑Pro: Run your standard prompts and compare total cost and latency against your current model; document where the 75% discount meaningfully lowers spend.
4. Lock down API keys and budgets: Rotate any public keys, switch to service‑account or newer key formats where possible, and set low spending alerts before green‑lighting agent projects.
5. Pilot Claude’s CFO Skill Stack for reforecasting: Create a Claude Project, upload the last 6–12 months of revenue/expense data, and run a monthly reforecast with the built‑in audit trail.