Healthcare AI sits in pilot because each new ePHI connection needs its own compliance build.
🔗 bit.ly/3QWIYq7
Connect AI moves the technical safeguards HIPAA requires to the platform level. The BAA is with CData.
English
CData Software
7.6K posts
CData Software
@cdatasoftware
CData is the data connectivity platform for AI applications, real-time analytics, and enterprise integration—access any data, anywhere.
Chapel Hill, NC Katılım Ağustos 2007
459 Takip Edilen2K Takipçiler
CData Software retweetledi
【基幹システム×HubSpot連携事例】
「事業ごとに顧客データがバラバラで、全社横断で活用できていない」
人材派遣事業を手がけるヒューマンリソシア株式会社様も、同様の課題を抱えていらっしゃったそうです。
派遣事業の基幹システムには顧客データが蓄積されていた一方、事業間でデータが分断され、全社横断での活用が難しい状態だったとのことです。
同社はCData Arcを導入し、基幹システムに手を加えることなく、蓄積データをCRM/SFAのHubSpotへ連携する仕組みを構築されました。
レガシーシステムの改修は大きな投資判断を伴いがちですが、既存環境はそのままに外側で連携層を作るという進め方は、情シス部門にとって現実的な選択肢の一つになるかもしれません。
結果として、工数60%削減に加え、事業を横断した顧客データ活用も実現されたとのことです。
レガシー連携の刷新を検討しているものの、社内にエンジニアリソースを割きにくい——という方には、参考になる事例かもしれません。
詳細はこちらの事例でご紹介しています 👇
jp.cdata.com/case-study/hum…
日本語
Connect AI's Universal Tools give agents a consistent 7-tool surface across hundreds of sources.
👉 bit.ly/4pbttr2
Workspaces scope data by function.
Toolkits control which tools each agent sees. Minimal surface, matched to the agent using it.
English
The fix is a different design question. Not what the API call does. What task the agent needs to accomplish. That reframing changes both the shape and count of what gets built.
💡 Most practitioners recommend staying under 10-15 active tools per agent.
English
The fastest MCP server to build is the one that wrecks your agent. Wrap each API endpoint as a tool, generate the manifest from the spec, ship by afternoon.
The mismatch doesn't show until that server meets two or three others.
English
CData Software retweetledi
When MCP tools are shaped around agent tasks instead of API endpoints, tool count drops, context stays lean, and the model spends its attention on the actual problem.
The blog covers what that design shift looks like in practice: bit.ly/4pbttr2
English
CData Software retweetledi
183,541 tokens at $0.596 per query. The same federated query across Salesforce, Snowflake, and ServiceNow with a scoped Custom Tool definition: 4,427 tokens at $0.027.
bit.ly/4oNDLxo 👉 Architecture is the cost lever, not prompt engineering.
English
You can have the best AI infrastructure in the world and a perfectly configured Databricks environment. If the pipeline from your on-prem SQL Server or legacy ERP is stale or broken, the whole thing stalls.
🔗 bit.ly/3PV1Mpl
English
Every control the NSA recommends is an independent engineering problem for teams running their own MCP servers.
CData Connect AI builds them in at the platform level.
Full breakdown of the advisory and what a secure deployment looks like: bit.ly/4y9Qe2w
English
The NSA's recommended controls: RBAC at the tool level, audit logging tied to session identity, sandboxed execution, cryptographic message signing, parameter validation, scans for unauthorized MCP servers.
None of these are protocol defaults.
English
MCP has no required authentication. That's a protocol-level design choice.
Authentication is left to implementors. Many skip it entirely.
The NSA's May 2026 advisory documents what happens next. 👇
English
CData Software retweetledi
「データ活用できない」の正体は、ツールでも戦略でもなく、蓄積を愚直に回し続けられていないことだと思っています。
私は結構ここをAIに助けてもらってます。SFAやCRMの提案をしていたのに、やっぱりデータ入力って苦手なんですよね。
それに、どこの組織でも最初はデータが無いし、データをうまく貯める構造も整っていない。データモデルだって、ビジネスに応じて変わっていく。
卵が先か、鶏が先か、みたいな話が付きないなと。
でも、データを入れる作業そのものをAIに手伝わせれば、インサイトへ辿り着くサイクルは一気に速くなると結構実感してます。
その活用イメージをCopilot #Cowork × #kintone × #CData Connect AI 記事第3弾で書きました。
jp.cdata.com/blog/cowork_ki…
日本語
CData Software retweetledi
Healthcare has been told to choose: AI-driven efficiency or HIPAA compliance.
CData Connect AI is now HIPAA-compliant.
👉 bit.ly/4aXgWBI
That trade-off no longer applies.
English
Your AI agent. Enterprise data. No connectors to build, no procurement cycle, no credit card.
👉 bit.ly/4vv1CEE
Connect AI Developer Edition is free MCP-native access to Salesforce, Snowflake, Workday, and hundreds more. Working agent in under 10 minutes.
English
Once the SQL is validated, there's nothing left for an MCP layer to broker. The agent writes a standalone JDBC app and ships it.
No agent at runtime, no model in the loop.
🔗 bit.ly/4oJrh9M
Build with AI. Ship without it.
English
Compute happens at the data layer, not in the LLM.
WHERE and ORDER BY get pushed to the source API. The model sees the handful of answer rows. Never the 20,000 raw records.
No LLM arithmetic mistakes.
English
AI agents are good at writing SQL. They're bad at learning Salesforce's SOQL, NetSuite's SuiteQL, and the API quirks of every other source they touch. CData CLI removes that problem.
A thread on the architecture. 👇
English