Together with
Hey builders,
This was the week AI got quieter — and more useful. One model refuses to write text. Its clones arrived fast. The practical question: can AI make a reliable call inside real software?
📦 From Our Partners
Learn AI in 5 minutes a day
You don't have to scroll every AI thread, track every new tool, or watch every demo.
The Rundown AI breaks it all down for you — the latest AI news, tools, and tutorials in one free 5-minute email every morning.
Trusted by 2M+ professionals at Apple, Google, and NASA.
Some teams never seem to stop moving. They're on Attio, the agentic CRM.
Every customer signal is captured in one shared context layer, always current and compounding. Agents and workflows build pipeline, chase every buying signal, and move deals forward, an always-on revenue engine running alongside your team.
With Attio, you’ll get:
Leads automatically prioritised and routed to the right rep
Expansion and risk signals caught the moment they land
Follow-ups written in your voice, already there when you arrive
Teams like Parallel, Turbopuffer, and Wordsmith build on Attio. Are you one of them?
🗺 THE WEEK IN 5
Jev shipped a smart
ifstatement for software → It returns typed choices, scores and confidence, not chatty text. Hype Scale: 🔨🔨🔨 (3/5) — promising; big performance figures remain vendor claims.The open Jev clone race began → Laya plus a new community benchmark gave builders an immediate alternative. Hype Scale: 🔨🔨🔨 (3/5) — momentum is real; calibration is hard.
AGENTS.md became the shared instruction sheet → Claude Code reads it when CLAUDE.md is absent. Hype Scale: 🔨🔨 (2/5) — boring infrastructure wins.
Pirate Face began preserving model weights → “Open” means little if a model vanishes on a company timetable. Hype Scale: 🔨🔨 (2/5) — verify every file.
Claude sped up biology models → Anthropic reports about 4× average speed-ups across 30+ models. Hype Scale: 🔨🔨 (2/5) — serious, but one company’s report.
📋 THE PROMPT
Turn any messy workflow into a decision map
Act as a workflow designer.
Task: [PASTE RECURRING TASK]
Inputs: [EMAILS / FORMS / DOCUMENTS]
Outcomes: [LIST 2–5 FIXED OUTCOMES]
Return only:
1. The one decision this task needs
2. Definitions for each outcome
3. Three edge cases for a human
4. A confidence threshold for auto-action
5. The smallest test set before automationWhy it works: It turns a fuzzy task into a small menu of outcomes. Let software own the rules; let AI supply the bounded judgment.
Make it yours: Try “triage new customer-support emails.”
🎨 FROM THE DRAFTING TABLE
📋 The full prompt:
Tiny autonomous lighthouse sorting glowing messages into three channels;
calm human behind glass; cinematic sci-tech, navy and amber, no text — 1200x800🔄 Make it yours: Swap {lighthouse} and {messages} for your workflow.
📺 WORTH WATCHING
Create your own voices with Gemini 3.8 text-to-speech → (1 min · Google DeepMind) — direct voice, pace and speaking style in plain language.
Why builders care: Output control is becoming an interface layer; consent and provenance must travel with the voice.
📐 ONE TERM A DAY
Today’s term: calibrated confidence
Plain English: “90% confident” is calibrated only if roughly nine of every ten similar 90% calls turn out right. It is a weather forecast that keeps its promises, not a model sounding very sure.
Why you should care: Confidence is not permission to automate. It is a signal for choosing when a human should review the hard cases.
📖 LOAD-BEARING READS
Why Jev is built to make decisions, not chat → — this reader-shared explainer walks through the “one forward pass” idea, the three output types, and why text generation is wasteful for a fixed yes/no decision.
Read it for the beginner-friendly mental model. For product claims and API details, keep the TypeSafe docs open beside it.
🎁 FREE SUPPLY DROP
Build your first agent without a course paywall. Microsoft’s open lessons cover tools, patterns and code exercises. → Grab it →
See every transformer part in working code. A free, readable build-from-scratch repository for understanding what sits under today’s models. → Grab it →
Start from real agent blueprints, not a blank tab. Open notebooks and patterns for multi-agent, RAG and research workflows. → Grab it →
1,000 Powerful Prompts for Claude AI — a prompt pack covering writing, coding, research and business work, so you stop starting from a blank box.
👷 NOW HIRING BUILDERS
Applied AI Architect @ forgd → — Remote USA; build Claude-native workflows with enterprise teams.
Applied AI Engineer @ Qualified Health → — Palo Alto hybrid; ship production healthcare-agent workflows.
Senior Full Stack AI Engineer @ Ombud → — Denver hybrid; own an agentic revenue-ops platform.
🧠 THE INSPECTION
Which Jev primitive chooses from a fixed list?
What shared file can Claude Code now read when CLAUDE.md is absent?
What should a system do when confidence falls below its safe threshold?
Answers, upside-down: 1) ǝɔᴉoɥƆ · 2) ᗡW.S┴NƎ⅁∀ · 3) uɐɯnɥ ɐ oʇ ǝʇɐlɐɔs
💬 Reply with one word: do you know how many AI agents are running on your machine right now? YES / NO
Tools Down
Swati, ByteBuilders