Issue #26 | November 7, 2026 | Weekly Intelligence Briefing

Gen Digital built an AI news engine on AWS that searches the live web for $7 per 1,000 queries and attaches a source URL to what it drafts.

Its own engineers then published the cost that isn't on the rate card. Reviewing AI output took 20 to 30 percent of their time.

That review line runs through the week. Google's new survey analysis in Forms ships with no published accuracy score, and Tencent's new model leads a blind test Tencent ran itself.

In a survey of 400 executives, 48% of the insights their analytics agents generate lead to actions that improve the business. Amazon's new coding setup opens a draft pull request and leaves the merge to you.

The vendors priced the machine this week. The checking is yours to budget.

Below: 2 signals, the Hype Check Spotlight, and the 1 tool worth your Monday.

Source: AWS Machine Learning Blog

Gen Digital published how it built the AI news engine inside Norton Revamp, its professional reputation platform launched in March 2026. The engine runs on Web Search on Amazon Bedrock AgentCore, per the AWS Machine Learning Blog.

AWS prices AgentCore Web Search at $7 per 1,000 queries, consumption-based with no minimum fee, checked November 4, 2026. The models underneath bill on their own token rates.

The search returns cited sources with URLs and publication dates. AWS says it "operates entirely within AWS with zero data egress," which is the vendor's own claim about its own service.

The engineers report 3x faster feature implementation, 85 percent code coverage and 99.9 percent uptime since launch. A small group of engineers built the platform.

The cost they published: AI output review consumed 20 to 30 percent of their time. Posts publish only upon approval, per Gen Digital's press release.

The search meter is $7 per 1,000 queries. The review time is 20 to 30 percent of the engineers' time, and that's the number your build estimate needs.

→ Pick 1 recurring research task, such as brand or competitor monitoring, and run it on the $7 meter for a month.

→ Write 20 to 30 percent review time into any agent build estimate before you approve it.

→ Keep a human approval step before anything publishes. Gen Digital's posts go out only on approval.

Read the full signal.

Source: Google Workspace Updates

Google announced Gemini insights in Google Forms on November 5. A Generate insights button in the Responses tab has Gemini write and run Python code against your actual responses.

Rapid Release domains started November 2, with up to 15 days for full visibility. Scheduled Release domains follow from November 19.

It returns 3 blocks: an Executive Synthesis, an Advanced Analysis and Correlations section, and Recommended Next Steps.

It needs at least 12 responses and 3 questions, and quiz forms are excluded. The older summarize feature works between 3 and 200 responses on text questions.

Google publishes no accuracy score for this feature. The Forms help center carries Google's standard caution that Gemini features may suggest inaccurate or inappropriate information.

It's included in Business Standard, listed at $14 per user a month with an annual commitment, checked November 6. Business Plus is $22, and Business Starter doesn't get the feature.

Google is showing up to 50% off the first 3 paid months for new Workspace customers, after a 14-day trial. At the full 50%, Business Standard comes to $7 per user for those months.

The counts come from your real responses. Deciding what they mean stays with you, and Google gives you no accuracy number to lean on.

→ Run Generate insights on your next feedback form with 12 or more responses, and check 3 of its counts against the raw rows.

→ Test it on a live form inside the 14-day trial before you commit to the annual plan.

→ Keep scored quizzes in Sheets. The feature refuses them.

Read the full signal.

Source: Tencent

Tencent launched Hy4, its open-source model, on November 4 with visual understanding and computer use for the first time. Its headline result comes from a blind test Tencent ran itself.

Community adoption. Hy4 is listed on OpenRouter with a 1 million-token context window, served by Tencent Cloud. That makes it reachable with an existing API key and no Tencent account.

Pricing model. Tencent held the Preview's rates: about $0.90 per million input tokens, $2.70 per million output and $0.045 per million cached input. Tencent Cloud TokenHub runs Hy4 at 50% off through December 31, 2026.

At the standard rate, 10 million input tokens and 2 million output tokens cost about $9 plus $5.40, about $14.40. The output side is where agent-heavy workloads spend most.

Benchmark data. In a blind evaluation with Tencent Financial Technology, 213 financial professionals scored Hy4 at 3.04 across 579 tasks. DeepSeek-V4.1-Flash-1017 scored 2.97, and Kimi K3.1 Pro and GLM-5.3 scored 2.68.

Tencent's internal evaluations put hallucination at 7.2%, down from 12.95% in Hy4 Preview.

Expert sentiment. The strongest independent number comes from financial AI company Rabyte, which measured a 12.03 percentage-point improvement over the Preview on the Investment Research Agent Benchmark. The office benchmarks are vendor-run.

Release maturity. Tencent says model weights and system cards follow "in the coming days." No public Hy4 weights page was live as of November 6.

A cheaper Hy4 Flash is on the way, with no published price.

The verdict: Hy4 leads DeepSeek by 0.07 points on a test Tencent ran itself, and its price is the stronger case.

Run 20 of your own tasks through it at the discounted rate before you move a workflow.

Read the full signal.

Source: Bloomberg Tech

ElevenLabs is an affiliate partner, and it gets the same treatment as a free pick. CEO Mati Staniszewski told Bloomberg Tech that Trainline rebooks tickets with a voice agent over 5 times faster than the old flow.

Revolut runs in-app support on the same stack. Extra agent minutes cost $0.08 on the vendor's rate card, and the free plan carries 15 agent minutes.

The weak spot for a business: audio made on the free plan can't be used for any business purpose. Commercial use starts on the $6 a month Starter plan.

Billing Warning. All ElevenLabs products draw from one credit pool, and unused credits are forfeited when you downgrade or cancel.

Refunds apply only within 14 days of payment, and only if no credits were used. Auto-renewal is on by default.

Pick the 1 queue your team answers from a script, run it on a voice agent for 2 weeks with disclosure on, and measure resolution time against your human baseline.

Read the full signal.

Agents that act on their own, and the people left to check them.

Only 48% of analytics agents' insights lead to actions that improve the business.
That's up from 37% a year earlier, in a survey of 400 executives. They blame a lack of business context (55%) and fragmented or inaccessible data (45%).
Reported by MIT Technology Review.

Amazon's AgentCore runs Claude Code in your AWS account, and the merge stays with you.
Each task runs in an isolated microVM and comes back as a draft pull request. The reference deployment's default $200 monthly Amazon Bedrock budget emails you without capping spend.
Announcement from AWS.

Meta's Spaces turns Muse agent chats into trackers that update from connected apps.
Muse agents had 10 million users in the U.S. and Canada within 6 weeks of launch, per Meta's earnings call. A tracker that syncs in the background can carry an error forward as fast as an update.
Reported by TechCrunch.

More than 800 Artex hacking tool deployments sat on the open web with no password.
Censys counted 1,465 hosts running an exposed ARTEX console as of November 4. Microsoft watched multiple incidents where initial targeting reached privileged access in under 3 hours.
Reported by TechCrunch.

The full week's signals, detailed breakdowns, and action items are on the site. If this issue earned its place in your inbox, forward it to whoever signs off on your AI budget.

This issue went out to subscribers Saturday. If you want next week's before it hits the web, subscribe at metadatamarketer.com/subscribe

Test. Cut. Share.

Moe Sbaiti, The AI Profit Wire https://metadatamarketer.com

Disclosure: some tools referenced in this newsletter are affiliate partners. Full disclosure and analysis at metadatamarketer.com.