TypeSafe's Jev Signals a Split Between "Chat" AI and "Decision" AI

TypeSafe's Jev Signals a Split Between "Chat" AI and "Decision" AI

ChatGPT co-inventor Diogo Almeida has come out of two years of stealth with Jev, a "decision model" from his startup TypeSafe AI, backed by roughly $40M in seed funding led by DCVC [4][5]. Unlike a conventional LLM, Jev doesn't converse — it returns structured choices, scores, or yes/no answers with confidence levels, and it's built to do this fast: claims of 20-200x speed gains and 40-400x cost reduction over frontier LLMs, with latency as low as 70ms [4][5][6].

The pitch is aimed squarely at production AI agents that need to route, classify, or make thousands of micro-decisions without burning frontier-model budgets on every call. Commentary around the launch frames Jev as infrastructure for reliability at scale — the unglamorous but essential layer that keeps agentic systems fast and affordable once they leave the demo stage [6].

It's a useful reminder that not every AI problem needs a chatbot. Sometimes you just need a fast, cheap, confident "yes."

Deel's Akai Platform Pushes Agentic Automation Into Back-Office Operations

Deel has launched Akai, an AI agent platform for operations that grew out of the company's own internal automation efforts [7][8]. Akai watches how workflows actually get done, then auto-builds and iterates automations across systems — voice, documents, payments — with humans stepping in only when agents hit a wall [7][9].

The internal numbers are the headline: Deel says its own deployment handles 100,000+ cases automatically per month, saves 91,000+ hours monthly, and runs 10,000+ live agents across finance, payments, and ops [8]. Pricing starts at $125/month per agent plus token usage, and early enterprise chatter points to thousands of agents already contributing meaningfully to ARR without corresponding headcount growth [9].

Akai is another data point in the broader shift from "AI helps you write faster" to "AI just does the operational work" — a trend worth watching for anyone whose job involves repetitive cross-system coordination.

What This Means For Your Meetings

Today's stories all point toward the same underlying shift: AI is moving from passive assistant to active infrastructure that captures, decides, and acts — and meetings sit right in the middle of that pipeline. Granola's traction shows that transcription quality alone is no longer a differentiator; what matters is what an AI system does with that transcript afterward — surfacing decisions, building context, and making it queryable weeks later [1][2].

Jev and Akai both point to where that "afterward" is heading. A decision model like Jev could plausibly sit downstream of a meeting transcript, turning discussion into fast, structured action items or routing decisions without waiting on a full LLM round-trip [4][5]. And Akai's model of watching workflows to auto-build automations is a preview of what meeting intelligence platforms will need to do next: not just recall what was said, but trigger what happens because of it [7][8].

For teams building a personal or organizational knowledge base from meetings, the lesson is clear — capture is table stakes, but the real value is in the layer that turns captured conversation into searchable, actionable knowledge across your entire meeting history. The tools winning attention right now aren't the ones with the best microphone; they're the ones that make yesterday's conversation useful today.

Key takeaway: The meeting is becoming the input, not the output — and the platforms that win will be the ones that turn conversation into structured, retrievable, actionable knowledge automatically.

Sources

1. https://docs.granola.ai/help-center/taking-notes/transcription
2. https://www.granola.ai/blog/granola-google-meet-integration-recording-transcription
3. https://techcrunch.com/2025/05/14/ai-note-taking-app-granola-raises-43m-at-250m-valuation-launches-collaborative-features/
4. https://aisocratic.org/news/chatgpt-co-inventor-diogo-almeida-launches-jev-a-decision-model-he-says-is-20200x-faster-than-frontier-llms
5. https://lukeocodes.dev/jev-system-one-model
6. https://www.therundown.ai/news/typesafe-jev-ai-decisions-software
7. https://www.akai.run/
8. https://www.deel.com/blog/akai-by-deel/
9. https://ecommercenews.uk/story/deel-launches-akai-to-automate-back-office-workflows