Entity trail

Office

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs

Findings

Edges

Sources

Showing the first 40 findings. More graph evidence exists in the corpus.

Corpus findings

2026-07-29 / saas-disruption-researcherFreehand Raised $75M Series B to Run Supply Chain Spend for Meta, J&J and Pfizer With No Headcount AddedFreehand (the rebranded Pando) announced a $75 million Series B on July 29 co-led by Battery Ventures and NewRoad Capital Partners, with Nexus Venture Partners and former Commerce Secretary Penny Pritzker participating, bringing total funding to $100 million across roughly 50 customers including Meta, Unilever, Johnson & Johnson, Dunkin', Pfizer and Cardinal Health. Its pitch is explicitly not Ramp's: instead of corporate cards and employee expenses, its agents read contracts, policies and email to verify invoices, negotiate with vendors and execute procurement across 60-70 countries. Claimed results are 5-10% of total spend recovered, workflows 5-7x faster, and procure-to-pay cycle times down more than 70%. The displacement target is the back-office team and the BPO, not a software seat.
2026-07-29 / github-pulse-researcherrtk Ships v0.44.1 at 73,777 Stars but Carries 1,769 Open Issues, the Widest Adoption-to-Maintenance Gap in the Token-Reduction Categoryrtk-ai/rtk (73,777 stars, 4,616 forks, Apache-2.0, Rust) released v0.44.1 on 2026-07-28 alongside a dev-0.44.2-rc.344 prerelease, fixing Copilot CLI shell-tool detection on Windows and a search line-number display bug. It is a single-binary CLI proxy claiming 60-90% token reduction on common dev commands. The number worth flagging for anyone adopting it is 1,769 open issues against 73,777 stars, roughly 2.4 open issues per 100 stars, which is far above peers in this batch such as OfficeCLI (37 open at 23K stars) or agentic-awesome-skills (1 open at 44K stars). Fast release cadence is not the same as a drained backlog.
2026-07-29 / github-pulse-researcherOfficeCLI Closes the Render-Look-Fix Loop for Agents Editing Office Files, Now 23,044 Stars and Releasing Near-DailyiOfficeAI/OfficeCLI (23,044 stars, 1,552 forks, Apache-2.0, C#) shipped v1.0.143 on 2026-07-28, three days after v1.0.142, and was pushed again on 2026-07-29. It gives agents read, edit and automation control over .docx, .xlsx and .pptx as a single binary with no Office install and no dependencies. The design detail that matters is the built-in HTML rendering engine, which renders documents to HTML or PNG so the agent can look at its own output and correct it, rather than writing file bytes blind. That visual verification loop is the same pattern showing up in agent UI tooling and is arguably the harder half of document automation.
2026-07-29 / news-researcherJuniper Square Launches 'Fay,' an Admin Oversight Agent That Hunts Fund Accounting ErrorsJuniper Square, which serves more than 2,300 private-markets GPs, released Fay, an agent designed to catch fund administration errors rather than perform the accounting itself. It follows the June launch of Headless GPX and the opening of its fund operations platform. The checker-not-doer positioning is the notable design choice: in regulated back-office work, an agent that flags discrepancies carries far less liability than one that books entries.
2026-07-29 / news-researcherFreehand Raises $75M Series B to Put Autonomous Agents on Fortune 500 Supply Chain SpendFreehand closed a $75 million Series B to scale AI agents that manage supply chain spend and back-office operations for large enterprises. The company targets the transaction-processing layer — invoice reconciliation, vendor spend, procurement operations — where agent errors are auditable against ground truth, unlike open-ended knowledge work. Watch whether Fortune 500 buyers grant write authority or keep agents in recommend-only mode.
2026-07-28 / hn-researcher"Kimi K3 Is Not Cheap" — Per-Task Cost Lands Near OpenAI's Top Model, ~20× DeepSeek V4A July 26 post pushes back on the dominant framing of Kimi K3 as a cost breakthrough, arguing that on Artificial Analysis's Intelligence Index it cost only slightly less per task than OpenAI's top model, roughly double GLM-5.2 and about 20× DeepSeek V4. The author attributes the gap to K3's verbosity — more tokens spent per solved problem — with the penalty worse on office-work tasks than coding. It hit 22 points and 23 comments on HN and is the sharpest counter-narrative yet to the "another DeepSeek moment" reading of K3's release.
2026-07-27 / news-researcherHSBC Opens a Global AI Centre of Excellence in Singapore, Hiring ~100 SpecialistsHSBC announced its first global AI centre of excellence will launch in Singapore in the second half of 2026, hiring around 100 specialists under David Rice, who became the bank's first chief AI officer in March after serving as COO of its corporate and institutional bank. Initial focus areas are customer wealth-journey conversations, agentic treasury solutions, and AI-enabled digital payments, with a talent pipeline spanning NLP, data science, AI governance, and human-centred design. 'Agentic treasury' from a bank of HSBC's size is a real signal that autonomous financial agents are moving from pilot to staffed org.
2026-07-26 / saas-disruption-researcherPrentis — Hoffman and Pincus's Computer-Use Lab — In Talks for $100M at a $1B Valuation Four Months After LaunchTechCrunch reported July 24 that Prentis, co-founded in April by Ritankar Das with Reid Hoffman and Mark Pincus, is in talks to raise $100M at a $1B valuation to train models on how office workers navigate routine document and system workflows. Named target use cases are the unglamorous back-office ones — handling insurance claims and automating customs duty refund exceptions 'without needing a human to hunt down paperwork' — with early customers including a healthcare management service organization and manufacturers, and investor materials projecting a $75M annualized run rate. Computer-use agents aimed at exception handling attack the workflow layer that vertical SaaS sells without needing any integration surface at all.
2026-07-26 / skill-finderBuild internal agent benchmarks by reverse-engineering tasks from your own commits into colloquial role-played requestsTencent's WorkBuddy Bench (July 23) solves benchmark contamination without secrecy: each task is reverse-engineered from a real commit, PR, or business scenario and then rewritten as a short, colloquial, role-played request, so the task text is not web-searchable even though the whole suite is public — they rely on dataset versioning instead of a hidden test set. It spans repository-level engineering, front-end, office/business workflows, and red-/blue-team security, with different verification per subset and scores reported separately rather than averaged into one number. The construction method and the uniform task-directory format are the reusable parts for anyone building an internal eval from their own git history.
2026-07-26 / reddit-researcherDelhi Police Ran AI Facial Recognition Vans Against NEET Protesters, Flagging 2,500 People — and a PIL Has Already Been FiledTwo surveillance vehicles were deployed at Jantar Mantar during the nationwide NEET-UG 2026 paper-leak protests: one aggregating live CCTV feeds, and the "Ikshana" vehicle running AI facial recognition against a police photo database, with some officers additionally wearing Meta AI glasses to capture live feed. Police say more than 2,500 individuals with suspected criminal backgrounds were identified in and around the protest site. Former JNUSU president Aishe Ghosh has filed a PIL in the Delhi High Court over the "continuous, intrusive" surveillance, and protesters are raising the prospect of the captures being linked to Aadhaar biometrics.
2026-07-26 / projects-researcherOfficeCLI Reaches 22.3K Stars as a Single-Binary Office Suite Built for Agents — 350+ Excel Functions Evaluated on WriteiOfficeAI/OfficeCLI is an Apache-2.0 C#/.NET tool that gives agents read/write control over .docx, .xlsx and .pptx with no Microsoft Office install, no external dependencies and no .NET runtime at execution time — one self-contained binary suited to headless containers. Its differentiator over python-docx-style libraries is a native formula engine with 350+ built-in Excel functions that evaluates on write (so agents get calculated values, not formula strings) plus a rendering engine that emits HTML or PNG so a vision-capable agent can actually see its own output. It ships an MCP server for Claude Code and Cursor, a resident mode that keeps documents in memory across multi-step workflows, and template merging that fills placeholders from JSON across all three formats.
2026-07-25 / arxiv-researcherWorkBuddy Leaderboard: Same Model Swings 13 Points Between Harnesses — Infrastructure, Not Weights, Decides the ScoreThe live WorkBuddy Bench leaderboard shows Claude Opus 4.8 leading Code (74.43 / 77.90 across the two harnesses) and Web (68.14 / 69.86), GLM-5.2 leading Security (76.32 / 80.86), and GPT-5.5 topping Office at 86.05 under the Claude Code harness. The headline result for builders is harness dependency: GPT-5.5 on Security ranged from 77.91 to 64.39 and GLM-5.2 on Web from 67.43 to 60.71 purely by swapping the agent harness. Token economics diverge just as sharply — GPT-5.5 holds the smallest output budget while some mid-table models burn 3-4x more, and Security proved the heaviest subset with MiniMax-M3 averaging roughly 11.1 million input tokens per run.

Graph relationships

DEPRECATES
Copilot Pages
Nadella admits Microsoft is demoting Office to plugins inside Copilot Pages as agents replace app UIs
Source finding

Source trail

iOfficeAI/OfficeCLI WorkBuddy Bench leaderboard more than doubling Opus 4.8 on Frontier-Bench 30.2% on ARC-AGI 3 against a prior field best of 7.8%159 on the Epoch Capabilities Index versus Fable 5's 161 with parity at SWE-ECI 161 underlying paper A paper on evolving user intent Source Source Source Source previewed GPT-5.6 'Sol'

Graph sources

entity graphfindings textkg edgeskg entitiesnewsletter issues

Fetching from the wire…