Slack Agents, OAuth, and OCR Updates #58

June 25, 2026 — 4 min read

Today's Letter

Anthropic introduced Claude Tag, a Slack-based way for teams to assign work to Claude inside shared channels
The beta is available now for Claude Enterprise and Team customers, and replaces the existing Claude in Slack app
Teams can grant @Claude access to selected channels, tools, data sources, and codebases, then assign tasks by tagging it in a thread
Anthropic says one shared Claude instance operates per channel, so multiple teammates can see progress and continue the same work context
Claude Tag can retain channel-specific context over time, while administrators keep memories, data access, and tools scoped by channel configuration
An optional ambient mode lets Claude post proactive updates, surface relevant information, and follow up on unresolved threads
Anthropic says Claude Tag can also run asynchronously over hours or days, and supports direct messages with personal tools and connectors
Administrators can set organization and channel token spending limits, review activity logs, and opt in to migration within 30 days
Anthropic says its internal version is already used heavily, with 65% of product-team code created through Claude Tag, and the launch model is Opus 4.8

Source: anthropic.com
More: techcrunch.com · fortune.com · siliconangle.com

GitHub added extended metadata support for Replicate secrets in secret scanning on June 23, 2026.
The update applies to the `replicate_api_token` secret type and surfaces richer context when leaked credentials are detected.
GitHub describes the change as an improvement to secret scanning rather than a new standalone security product.
The primary effect is better incident context for exposed Replicate API tokens, which may help triage and remediation workflows.
The changelog entry does not list pricing changes, rollout stages, or additional configuration requirements in the announcement.
GitHub links to its secret scanning documentation for the full supported-secret list and implementation details.

Source: github.blog

Microsoft Dev Blogs published a remembrance for Tony Krueger, described as the engineer who made Word's spell-check feedback non-blocking and added the red squiggles for misspellings
The post says he worked on Word 1.0, 1.1, 2.0, Word for OS/2, Word for Mac, and later returned for Word 6.0 and several later releases
Early Word spell checking required a manual pass or an idle-time background check that could still block foreground actions such as save and exit
Krueger's change made spell checking less intrusive by surfacing issues inline during editing instead of waiting for a dedicated review flow
The article notes that green squiggles for grammar followed later, and that red, green, and blue underline patterns are now common across word processors and other software
Raymond Chen says Krueger's work reached a broad user base despite his name being less widely known than the feature itself
The post was published on June 22, 2026 on The Old New Thing, framing the feature as an early Microsoft UI pattern that later became standard elsewhere

Source: devblogs.microsoft.com
More: news.hada.io

Mistral AI released Mistral OCR 4 on June 23, 2026 as a document OCR model for enterprise ingestion and retrieval pipelines.
The model returns extracted text with bounding boxes, block classification, and inline confidence scores.
It supports 170 languages across 10 language groups, including low-resource and specialized languages.
OCR 4 can run in a single container for fully self-hosted deployments aimed at data residency and compliance requirements.
Mistral positions it as an ingestion component for enterprise search, RAG, and domain-specific retrieval workflows.
API pricing is $4 per 1,000 pages, or $2 per 1,000 pages through the Batch API with a 50% discount.
Document AI in Mistral Studio uses the same engine and is priced at $5 per 1,000 pages.
Mistral says OCR 4 reached an 85.20 score on OlmOCRBench and averaged a 72% win rate in human preference tests.

Source: mistral.ai
More: news.hada.io · the-decoder.com

Cloudflare opened self-managed OAuth to all customers on 2026-06-24, extending delegated API access beyond manually onboarded partners.
Developers can now offer standard OAuth flows for SaaS integrations, internal platforms, and agentic tools instead of relying on API tokens.
Cloudflare updated consent screens, added dashboard-based revocation, and exposed app ownership more clearly to reduce OAuth phishing risk.
The platform previously used Hydra for Cloudflare OAuth, but growing developer-platform usage and agentic workflows pushed a major engine upgrade.
For the 1.X upgrade, Cloudflare rewrote SQL migrations to use CREATE INDEX CONCURRENTLY and built a custom Hydra version that avoided SELECT *.
Cloudflare planned the 2.X move as a blue-green migration because an in-place upgrade was impractical and the migration window could last multiple hours.

Source: blog.cloudflare.com
More: publish.my

Jocoletter curates AI, software, and product trends for developers and builders.

#Anthropic #Cloudflare #GitHub #Microsoft #MistralAI