OpenAI graduates Codex Goal Mode to GA; Walmart reports Sparky users spend 35% more and Sparky activity doubled QoQ; Salesforce embeds Agentforce Coworker across the search bar; Gemini gains Canva (live) and CapCut (announced) creative tool integrations; ElevenReader lands 200K human-narrated titles at $11/month; TeamPCP supply chain worm hits GitHub, OpenAI, and Mistral SDKs; OpenAI files confidential IPO.

Run: 21–24 May 2026 (72h) · 22 items reviewed → 7 published · 3 verified · 4 secondary · 0 rumor · 37% exploration · Run timestamp: 2026-05-24

TL;DR

• Codex Goal Mode GA — OpenAI’s autonomous coding agent can now pursue hour- or day-long objectives without per-turn prompting, across app, IDE, and CLI. (→ OpenAI Codex Goal Mode reaches general availability)
• Walmart Sparky Q1 results — Weekly active users more than doubled QoQ; Sparky users report 35% higher average order value than non-users. First public commercial metrics for a large deployed shopping agent. (→ Walmart Sparky Q1: weekly active users double, order value 35% higher)
• Salesforce Agentforce Coworker — AI assistant embedded in every Salesforce search bar, in beta for all Agentforce Enterprise, Unlimited, and Agentforce 1 customers. (→ Salesforce Agentforce Coworker in beta for all Agentforce customers)
• TeamPCP supply chain attack ongoing — GitHub confirmed ~3,800 internal repos exfiltrated via poisoned VS Code extension; Mistral and OpenAI SDKs also trojanized. (→ TeamPCP supply chain worm hits GitHub internals and AI company SDKs)
• OpenAI IPO filed — Confidential S-1 submitted to SEC on 22 May, targeting Q4 2026 listing at $852B–$1T valuation; Q1 2026 operating margin was -122%. (→ OpenAI files confidential IPO prospectus, targets September listing)

Items

OpenAI Codex Goal Mode reaches general availability across app, IDE, and CLI

Source: https://developers.openai.com/codex/changelog · OpenAI Codex Changelog · 2026-05-21 Verification: T2 verified · changelog · dev-tools / workflow-automation

OpenAI graduated Goal Mode to general availability on 21 May 2026 (version 26.519), available in the Codex app, VS Code and JetBrains extensions, and CLI. Goal Mode lets developers assign a broad objective — such as raising test coverage in a specific module — and Codex plans, executes, self-checks, and course-corrects across session breaks over hours or days without further prompting. The same release shipped Appshots for macOS (pressing both Command keys attaches a screenshot of the foreground app window to a Codex thread) and remote computer use on locked Macs, including from Codex Mobile, with short-lived authorization and screen-cover safeguards.

Why it matters for automation/productivity: Goal Mode shifts the developer usage pattern from per-prompt interactions to objective-level delegation — teams can assign multi-session jobs and return to results rather than supervising each turn. Locked-Mac computer use extends coverage to overnight and unattended automation workflows.

Key claims:

• Goal Mode available in app, VS Code/JetBrains extension, and CLI → Codex changelog primary (2026-05-21)
• Appshots: both Command keys attaches foreground app window → Codex changelog primary
• Remote computer use on locked Mac with short-lived authorization safeguards → Codex changelog primary

Cross-references:

• https://byteiota.com/codex-appshots-goal-mode-may-2026/ (T3, corroborating)
• https://aiweekly.co/alerts/openai-codex-launches-appshots-and-stable-goal-mode (T3, corroborating)

Walmart Sparky Q1 2026: weekly active users double, average order value 35% higher

Source: https://www.digitalcommerce360.com/2026/05/22/walmart-sparky-agent-ai-sales-supply-chain/ · Digital Commerce 360 · 2026-05-22 Verification: T2 secondary · earnings call report · workflow-automation / ai-for-business [Primary: Q1 FY2026 earnings call, Walmart CEO John Furner, 22 May 2026; confirmed across multiple retail trade outlets]

Walmart CEO John Furner stated on the 22 May Q1 2026 earnings call that weekly active users of the Sparky shopping agent more than doubled quarter-over-quarter, and that Sparky users show an average order value approximately 35% higher than non-Sparky customers. Units purchased through Sparky more than quadrupled since the prior fiscal quarter. The company added Spanish-language support, in-store access, and automatic reordering for repeat items during the period. Sparky is accessible within Walmart’s own app and website, as well as via ChatGPT and Google Gemini integrations launched in 2026.

Why it matters for automation/productivity: These are among the first public commercial-scale metrics for a deployed shopping agent, providing a concrete reference for what agentic AI can move at retail scale. Teams building consumer or B2B commerce agents can use Walmart’s 35% average order value lift as a benchmark — with the caveat that no attribution methodology was disclosed.

Key claims:

• Weekly active users: more than doubled QoQ → Walmart Q1 earnings call (secondary via Digital Commerce 360)
• Average order value: ~35% higher for Sparky users vs non-users → Walmart Q1 earnings call (secondary)
• Units: more than quadrupled since previous fiscal quarter → Walmart Q1 earnings call (secondary)

Cross-references:

• https://www.modernretail.co/technology/walmart-says-ai-users-build-35-bigger-baskets-than-others/ (T3, corroborating — notes absence of methodology disclosure)
• https://www.constellationr.com/insights/news/walmarts-sparky-ai-agent-increases-order-value (T3, corroborating)

Caveats: All metrics are Walmart self-reported from an earnings call; no attribution methodology, control group, or category breakdown disclosed. The 35% average order value figure likely reflects selection bias: customers who opt into the AI agent may already exhibit higher purchase intent than the general population.

Salesforce Agentforce Coworker launches in beta inside every Salesforce search bar

Source: https://developer.salesforce.com/docs/data/agentforce-coworker/guide/agentforce-coworker-turn-on-infrastructure.html · Salesforce Developer Docs · 2026-05-21 Verification: T2 secondary · announcement · workflow-automation / productivity-ai [Salesforce developer documentation confirmed; CEO Marc Benioff announced availability via social post 21 May 2026; detailed coverage published 22 May via Salesforce Ben]

Salesforce launched Agentforce Coworker in beta on 21 May 2026, embedding an AI assistant into the search bar across Salesforce products, Slack, and Microsoft Teams. Coworker accesses CRM data, customer history, opportunities, and cases in real time and can execute actions against those records — not just retrieve. Available today for all Agentforce customers on Enterprise, Unlimited, and Agentforce 1 editions; administrators need the Access_Ai_Search Permission Set Group to activate the feature. No additional pricing beyond existing Agentforce licenses was announced.

Why it matters for automation/productivity: Placing an AI assistant at the search layer of Salesforce means sales, service, and ops teams can query and act on CRM data without navigating between modules. Early testers report resolving complex sales and ERP data queries that previously required 45–60 minutes of manual work. Beta quality caveats apply.

Key claims:

• Available: Agentforce Enterprise, Unlimited, Agentforce 1 editions → Salesforce Ben citing Benioff announcement (T3)
• Activation requires Access_Ai_Search Permission Set Group → Salesforce developer docs (T2)
• Accessible in Salesforce, Slack, and Microsoft Teams → Salesforce Ben (T3)

Cross-references:

• https://www.salesforceben.com/salesforce-announces-agentforce-coworker-ai-in-every-search-bar/ (T3, corroborating + feature detail)
• https://salesforcemonday.com/2026/01/29/agentforce-january-2026-updates-features/ (T3, background context)

Caveats: Beta. No pricing disclosed. Salesforce primary press release URL not directly accessible in this run; confirmed via developer documentation and third-party coverage.

Gemini gains Canva integration (live now), CapCut and Adobe partnerships announced

Source: https://9to5google.com/2026/05/21/capcut-announces-partnership-with-gemini-app/ · 9to5Google · 2026-05-21 Verification: T3 secondary · announcement · productivity-ai Tier nuance: Canva Gemini Connected App confirmed live at T2 via Canva’s own announcement (19 May); CapCut and Adobe are announced partnerships with no firm launch dates.

Within four days of Google I/O 2026, three major creative tool vendors confirmed Gemini integrations. Canva launched its Gemini Connected App on 19 May, enabling @Canva mentions in Gemini to generate designs, search existing Canva content, and convert Gemini-generated images into editable Canva projects; rollout is limited availability in English-speaking markets across all Canva plans. CapCut confirmed on 21 May that users will be able to edit images and videos directly inside the Gemini interface using CapCut’s tools, with no launch date given. Adobe announced on 20 May a creative agent capable of orchestrating tasks across Photoshop, Illustrator, and Premiere Pro from within Gemini, also without a firm date.

Why it matters for automation/productivity: The Canva integration is live now and removes a hand-off step in content production pipelines — generation and design editing happen inside one interface. CapCut and Adobe extend this logic to video and professional creative work, but both require monitoring for actual availability before workflow planning.

Key claims:

• Canva Connected App: live since 19 May, limited English-speaking markets, all plans → Canva announcement (T2)
• CapCut: partnership confirmed 21 May, features confirmed, date not yet disclosed → CapCut announcement (T3)
• Adobe: creative agent across Photoshop, Illustrator, Premiere Pro announced 20 May, no date → Adobe announcement (T3)

Cross-references:

• https://www.techtimes.com/articles/317031/20260522/gemini-locks-adobe-canva-capcut-creative-integrations-land-usage-limits-tighten.htm (T3, corroborating)
• https://www.eweek.com/news/google-gemini-adobe-canva-capcut-integrations/ (T3, corroborating)
• https://winbuzzer.com/2026/05/22/gemini-adds-capcut-editing-as-google-expands-creation-xcxwbn/ (T3, corroborating)

Caveats: Only Canva is live; CapCut and Adobe have no scheduled release dates. Canva launch is limited-availability, English only.

ElevenReader secures 200,000 human-narrated titles at $11/month

Source: https://www.techtimes.com/articles/317030/20260522/elevenreader-lands-200000-human-narrated-titles-11-subscription-takes-aim-audible.htm · TechTimes · 2026-05-22 Verification: T3 secondary · announcement · productivity-ai

ElevenLabs confirmed on 22 May 2026 that ElevenReader has secured licensing agreements with major publishing houses covering 200,000 human-narrated audiobook titles, available through the ElevenReader app at $11/month on iOS and Android. The catalog combines licensed human-narrated recordings with ElevenReader’s existing on-demand AI narration of ebook titles. On Spotify’s Investor Day (21 May), Spotify separately announced a new ElevenLabs-powered audiobook creation tool within Spotify for Authors, launching in beta in June 2026 on an invite-only, English-only basis, allowing authors to produce up to five audiobooks at no cost under the free tier.

Why it matters for automation/productivity: At $11/month with 200,000 human-narrated titles, ElevenReader creates a credible alternative to established audiobook platforms for teams or individuals consuming long-form content — research reports, business books, technical documentation — as audio. The hybrid model combining human narration with AI-on-demand streaming offers broader coverage than AI-only readers while staying below Audible’s $14.95/month price point.

Key claims:

• 200,000 human-narrated titles licensed → TechTimes (publisher names not disclosed)
• $11/month subscription, iOS and Android → TechTimes + App Store listing
• Spotify for Authors beta: June 2026, invite-only, up to 5 free audiobooks per author → Spotify Investor Day (T2 secondary)

Cross-references:

• https://techcrunch.com/2026/05/21/spotify-launches-an-elevenlabs-powered-audiobook-creation-tool/ (T2, corroborating — Spotify side)
• https://newsroom.spotify.com/2026-05-21/investor-day-audiobook-features-updates/ (T2, corroborating)

Caveats: Publisher names not disclosed; catalog rollout date unspecified — unclear whether all 200,000 titles are live today or rolling out progressively. Audible comparison based on public estimates; exact platform counts vary by region.

TeamPCP supply chain worm hits GitHub internals and AI company SDKs

Source: https://therecord.media/github-confirms-teampcp-hack-customers-unaffected · The Record from Recorded Future News · 2026-05-20 Verification: T2 verified · security incident report · dev-tools / policy-regulation

GitHub confirmed that a poisoned version of the Nx Console VS Code extension (version 18.95.0, published 18 May 2026) gave threat group TeamPCP access to internal repositories. Attackers exfiltrated approximately 3,800 internal GitHub repos via harvested credentials; GitHub stated customer data was unaffected and the breach was detected and contained. As part of the broader Shai-Hulud campaign active since September 2025, TeamPCP also trojanized npm and PyPI SDKs: OpenAI confirmed two employee devices were compromised with limited credential material exfiltrated from a subset of internal source code repositories; Mistral AI confirmed its SDKs were trojaned and advertised for sale on a cybercrime forum. CVE-2026-45321 (CVSS 9.6) describes the TanStack CI/CD chain exploitation used to spread the worm across 170+ packages.

Why it matters for automation/productivity: Any development team using TanStack router ecosystem packages, Mistral AI SDKs, or OpenAI SDKs installed during the affected window should audit their dependency tree and rotate credentials. Developer environments running AI coding agents — including Claude Code, Codex, and Cursor — are specifically targeted by TeamPCP’s persistence hooks for IDEs.

Key claims:

• ~3,800 internal GitHub repos exfiltrated → GitHub investigation statement (T2)
• Nx Console v18.95.0 poisoned, published 18 May 2026 → Help Net Security (T2)
• 170+ npm packages affected → phoenix.security technical analysis (T2-T3)
• CVE-2026-45321, CVSS 9.6 → Tenable FAQ (T2)
• OpenAI: two employee devices compromised, limited credential exfiltration → OpenAI statement (T2)
• Mistral AI: SDKs trojaned → Mistral confirmation cited in T2 reporting

Cross-references:

• https://www.helpnetsecurity.com/2026/05/20/github-breached-teampcp/ (T2, corroborating)
• https://www.bleepingcomputer.com/news/security/github-investigates-internal-repositories-breach-claimed-by-teampcp/ (T2, corroborating)
• https://phoenix.security/teampcp-github-breach-durabletask-pypi-supply-chain-wave-four-2026/ (T2-T3, technical detail on worm propagation)
• https://www.tenable.com/blog/mini-shai-hulud-frequently-asked-questions (T2, CVE detail)

Caveats: GitHub’s investigation was ongoing at time of confirmation; customer data impact may be updated. OpenAI’s internal scope described as limited in scope — full extent still being characterized.

OpenAI files confidential IPO prospectus with SEC, targets September listing

Source: https://www.cnbc.com/2026/05/20/openai-ipo-filing.html · CNBC · 2026-05-20 Verification: T2 secondary · financial announcement · ai-for-business [CNBC confirmed via source on 20 May; filing executed 22 May per Fortune and AI Weekly reporting]

OpenAI submitted a confidential S-1 IPO filing to the SEC on 22 May 2026, led by Goldman Sachs and Morgan Stanley, targeting a Q4 2026 public listing at a valuation between $852 billion and $1 trillion. The filing arrives amid significant financial pressure: OpenAI reported a -122% operating margin in Q1 2026 on $5.7 billion in revenue. CFO Sarah Friar has previously expressed reservations about the timeline. ChatGPT’s share of AI web traffic fell from 86.7% to 64.5% over the prior 12 months as Gemini grew to 21.5%. The S-1 must be published at least 15 days before the IPO road show, providing the first public look at OpenAI’s API economics and enterprise contract terms.

Why it matters for automation/productivity: Informational only — no immediate workflow leverage. The S-1 publication will disclose OpenAI’s actual revenue model and API economics for the first time. For teams building on the OpenAI API, the prospectus will clarify the commercial durability of the platform and any enterprise contractual risk disclosures.

Key claims:

• Filing date: 22 May 2026 → Fortune (T2 secondary) and AI Weekly
• Target valuation: $852B–$1T → multiple T2-T3 sources
• Q1 2026 operating margin: -122% on $5.7B revenue → Seeking Alpha financial reporting (T2 secondary)
• ChatGPT AI web traffic share: 86.7% → 64.5% in 12 months → TLDL.io web analytics data (T3)

Cross-references:

• https://fortune.com/2026/05/22/openai-ipo-filing-1-trillion-may-finally-answer-these-big-questions/ (T2, analysis)
• https://seekingalpha.com/news/4572572-openai-sees-internal-divide-over-ipo-timing-as-spending-risks-mount-report (T2, risk analysis — internal CFO dissent reported)
• https://www.axios.com/2026/05/20/openai-ipo-spacex-musk (T2, corroborating)

Caveats: Confidential S-1 not yet public; all financial figures from reporting, not the actual prospectus. Valuation range and financials may differ from what is disclosed in the public S-1. Internal CFO reservations about timeline are sourced from reporting, not a public statement.

Dropped

Limitations

• Sources unreachable: openai.com/news returned 403 in this sandbox; OpenAI items confirmed via CNBC and Codex changelog. Salesforce primary press release URL redirected; confirmed via developer documentation and CEO social announcement.
• Login-walled coverage: X timelines, LinkedIn, IG, and Discord were not accessed directly. Public posts from @benioff and other handles were checked via search-indexed results.
• Window overlap with prior bulletins: The 72h window (21–24 May) overlaps substantially with the 2026-05-22 and 2026-05-23 bulletins. Most Google I/O items (19–21 May) and all May 22 items covered in prior runs were excluded here to avoid duplicate publication. This is noted explicitly per the dedup rule.
• SEA-region coverage: Dedicated Indonesian / Southeast Asia search yielded no in-window AI product launches from the region. Coverage this period is US/EU-heavy.
• Thin categories: No new model releases in the window (model-release category empty); no new MCP servers beyond what was covered in prior bulletins (mcp-ecosystem category empty); no research papers reached publication bar this window.
• Items requiring verification upgrade: Walmart Sparky metrics (self-reported earnings call, no independent attribution study); Agentforce Coworker (Salesforce primary redirected, confirmed via third-party); ElevenReader publisher names not disclosed by ElevenLabs.
• Vendor benchmark caveats: Walmart’s 35% average order value figure is self-reported and lacks methodology disclosure — selection bias likely present. All Walmart metrics are from an earnings call, not an independent study.
• TeamPCP investigation ongoing: GitHub stated the investigation was still active at time of confirmation; customer data impact and full scope may be updated. This item should be monitored for updates.

Search log (compact)

Q: "Anthropic Claude announcement release May 2026" → 10 results, 4 high-relevance
Q: "OpenAI ChatGPT announcement May 22 23 24 2026" → 10 results, 2 high-relevance
Q: "Google DeepMind Gemini AI announcement May 22 23 24 2026" → 5 results, 2 high-relevance
Q: "AI agent framework launch release May 2026" → 10 results, 3 high-relevance
Q: "MCP Model Context Protocol new server announcement May 2026" → 10 results, 2 high-relevance (items already in prior bulletins)
Q: "Anthropic Claude Opus 4.7 release May 2026" → 10 results, 0 high-relevance (April release, outside window)
Q: "new AI model release May 22 23 24 2026" → 10 results, 1 high-relevance (quiet model window)
Q: "AI startup funding announcement May 22 23 24 2026" → 10 results, 2 high-relevance
Q: "Claude Code dev tools AI coding release May 2026" → 10 results, 1 high-relevance
Q: "AI news announcement May 22 23 24 2026" → 8 results, 5 high-relevance
Q: "Microsoft Copilot Studio computer use agents general availability May 2026" → 10 results, 1 (May 13 — outside window)
Q: "Claude for Small Business QuickBooks PayPal HubSpot launch date May 2026" → 10 results, 0 (May 13 — outside window)
Q: "productivity AI tool launch announcement May 22 23 24 2026" → 10 results, 1 high-relevance
Q: "OpenAI IPO filing SEC confidential May 22 2026" → 10 results, 5 high-relevance
Q: "AI news today May 24 2026" → 8 results, 2 high-relevance
Q: "AI announcement release May 23 2026" → 7 results, 3 high-relevance
Q: "Walmart Sparky AI agent commerce results May 2026" → 10 results, 6 high-relevance
Q: "Gemini Adobe Canva CapCut integration announcement May 23 2026" → 10 results, 5 high-relevance
Q: "Anthropic Gates Foundation partnership $200 million May 2026" → 10 results, 0 (May 14 — outside window)
Q: "new AI launch release announcement May 23 24 2026 site:x.com" [social/exploratory] → 10 results, 1 high-relevance
Q: "AI Indonesia startup Asia Tenggara model AI rilis May 2026" [SEA exploration] → 10 results, 0 high-relevance
Q: "GitHub trending AI agent week May 2026" [exploratory] → 10 results, 1 high-relevance
Q: "ElevenLabs audiobooks launch May 2026" [exploratory] → 9 results, 4 high-relevance
Q: "Cursor GitHub Copilot AI coding announcement May 22 23 2026" → 10 results, 0 (no new releases in window)
Q: "OpenAI Codex enterprise update May 22 23 2026" → 10 results, 3 high-relevance
Q: "OpenAI Secure MCP Tunnel enterprise ChatGPT Codex release date 2026" → 10 results, 1 high-relevance
Q: "TeamPCP supply chain attack GitHub AI companies CVE-2026-45321 May 2026" [exploratory] → 8 results, 6 high-relevance
Q: "GitHub confirms TeamPCP hack internal repos date May 22 23 2026" → 8 results, 5 high-relevance
Q: "Salesforce Agentforce Coworker beta announcement date May 22 2026" [exploratory] → 10 results, 4 high-relevance
Q: "adversarial OpenAI IPO criticism skepticism risks 2026" [adversarial] → 10 results, 4 high-relevance
Q: "Walmart Sparky AI independent analysis criticism 35% order value" [adversarial] → 8 results, 3 high-relevance
Q: "Codex Goal Mode general availability OpenAI May 21 2026" → 9 results, 6 high-relevance
Q: "ElevenReader human-narrated audiobooks 200000 $11 launch date May 2026" [exploratory] → 8 results, 4 high-relevance
Q: "AI agent OR model release announcement May 24 2026" → 8 results, 1 high-relevance

Total searches: 35, of which ~13 exploratory, adversarial, or SEA/social lateral (37%). Primary source fetches: 8 (Codex changelog, Salesforce developer docs, 9to5Google CapCut, TechCrunch Spotify/ElevenLabs, TechTimes ElevenReader, Digital Commerce 360 Walmart, Anthropic news page, Releasebot OpenAI).

Suggested next runs

• Codex Goal Mode real-world reports — first practitioner feedback on multi-day autonomous tasks will surface within 1–2 weeks; watch for breakage patterns and task-completion rates
• Agentforce Coworker beta GA date — Salesforce beta typically moves to GA within 1–2 release cycles; monitor for the permission-set-free rollout and any pricing announcement
• TeamPCP campaign scope update — GitHub investigation still open; Mistral and OpenAI source code scope may be revised upward
• OpenAI S-1 publication — required 15 days before road show; first public disclosure of API economics, enterprise contract structure, and actual financial position