🛸👽 Dr. David Jacobs on Alien Abduction Hierarchies - Insectoids at the Top? ⚠️🌌
My CppCon keynote went really well!
Fun times talking about new memory hierarchies and the best ways we can programmatically prepare for them in advance.
Tesla and SpaceX operate on a fundamentally different principle than many traditional companies:
Minimize the distance between the people who know what's happening and the people who can do something about it.
Traditional corporate hierarchies emphasize control, reporting relationships, and clearly defined authority.
Musk emphasizes engineering, speed, direct communication, and accountability for results.
The advantage isn't simply having fewer managers.
It's having the shortest path for truth.
Show more
At SEMICON Taiwan’s imec Technology Forum, memory architecture and 3D integration emerged as closely connected parts of this challenge. Gouri Sankar Kar examined next-generation memory hierarchies for edge, physical, and agentic AI, while Zsolt Tőkei addressed the interconnect and integration technologies needed for future computing systems. A common direction was clear: select memory according to how data is used, then connect it vertically at the density and proximity the workload requires.
Show more
“Never be distracted by people's glamorous portraits of themselves and their lives; search and dig for what really imprisons them.”
- Robert Greene
Robert Greene is an American author widely recognized for his writing on power, strategy, seduction, human nature, and social psychology. His work examines how people pursue their goals, conceal their motives, react to status, and navigate relationships, competition, and social hierarchies.
Greene gained worldwide recognition with The 48 Laws of Power, which draws on historical figures—including rulers, political leaders, strategists, and other influential personalities—to examine how power is gained, maintained, challenged, and lost.
Show more
Introducing the first in our Parsed by LlamaParse series. We're outlining the stakes of 'getting parsing wrong' in consequential documents.
We’re starting with the U.S. Energy Information Administration’s September 2026 Short-Term Energy Outlook, a dense government report covering energy supply, demand, prices, and forecasts.
Table 7a alone packs multiple years, quarters, row hierarchies, units, and footnotes into one electricity-industry table.
Take 1,186.
Parsed correctly, it means electricity sales to ultimate customers in Q3 2026, measured in billion kilowatthours.
Parsed incorrectly, it could be assigned to the wrong quarter, metric, or unit , which means the error can flow straight into a dashboard, forecast, alert, or AI application.
And the values aren’t the only thing that matters. Footnotes define the data too: “small-scale solar,” for example, refers to systems under one megawatt, not solar generation overall.
LlamaParse preserves the structure and context that make document data usable downstream: whether you’re populating a database, updating a dashboard, running forecasting workflows, or building an AI app over complex documents.
Source doc here:
Show more
Spreadsheets are great for storing numbers. They’re terrible for showing them off. 📈👀
Lately, the Stitch community has been using their design partner to turn raw data dumps into sleek, production-ready dashboards:
🔹 Paste in CSVs, analytics exports, or workout logs
🔹 Ask Stitch for a specific vibe, layout, or design system
🔹 Instantly explore high-fidelity charts, stat callouts, and clean data hierarchies
🤖 Automation Hack: Pair Gemini Spark or Antigravity with the Stitch MCP Server. You can automate your daily tracking so a fresh, tailored dashboard is generated before your first cup of coffee.
Check out the demo to see how we turn raw data into eye candy! 🔊👇
Show more
🤔 "Now that AI can write code, does software design even matter anymore?" A lot of people are starting to feel that way. This article argues the exact opposite is true.
Coding is just one of five activities in software development — alongside requirements understanding, design, verification, and support. Human developers can get by with a slightly vague design because they lean on team tribal knowledge and experience. AI coding agents (AI-SDEs) can't do that. Without design information made explicit, they miss chances to reuse existing code, quietly generate duplicates instead, and every subsequent fix to a mature system compounds the complexity further.
What's even more interesting is the claim that the economics of design itself are shifting. Object-oriented design and class hierarchies were built to reduce the amount of code a human has to write. But once AI implementation cost approaches zero, a design that simply explains behavior clearly to the AI can beat an elaborate reuse hierarchy. The question flips from "how do we minimize code written" to "what design produces the best software when an AI implements it."
Software Design in the Age of AI: Why AI coding makes software design more important
The faster and cheaper AI makes writing code, the more the quality of the design behind it seems to matter.
#
SoftwareDesign# #
AIAgents#
Show more
Fun fact: In the early days of SpaceX, Fridays had a different sound, or rather, taste 😋
The company had a wonderful tradition: after a week in which everyone was making the impossible a reality, on Fridays, SpaceX would order ice cream for all employees of the company, paying for the order with a corporate card.
The interesting thing was that no matter what position you held, whether you were a rookie or a vice president, you would go get ice cream for everyone and bring it to the office.
Later, a tradition arose to stay in the office after work to play Quake, where employees could beat each other or Elon Musk, who would set impossible tasks for them all day and demand their completion🤣 All positions and hierarchies remained behind the door, in the game, everyone was equal.
That's how SpaceX and Elon entertained their people. That's how they brought the team together, with ice cream, a game, and the feeling that everyone is equal.
Show more
Harness Engineering Practices
P4. Contract Subagents as Distillers
🎯 Point
If a subagent returns raw logs from 50 files to the parent, the parent's context is instantly dead. Return conclusions, not transcripts.
📝 Overview
Define the subagent's contract as "return conclusions, not minutes." Explore 50 files, return 5 lines of findings. This isn't courtesy — it's memory management discipline for context hierarchies.
🔍 Explanation
The primary purpose of using subagents is to protect the parent agent's context window. But if subagents return raw exploration results as-is, that purpose is entirely defeated. Subagents should function as "distillers" — consuming large volumes of information and returning only the essence. By explicitly defining this contract in the prompt and constraining the response format, you maintain efficiency across the entire context hierarchy. This also aligns with token economics: use cheap models for broad exploration, expensive models for deep judgment.
🛠 How to Practice
- Explicitly state "return conclusions, not transcripts" in subagent prompts and constrain response format (line limits, structure)
- Design token economics: use cheap models for exploration tasks, expensive models for parent agent judgment
- Monitor subagent response sizes and add guards that request re-distillation when thresholds are exceeded
- Adjust distillation granularity per task (3 lines for classification, 10 lines for analysis, etc.)
💼 Use Cases
- Large codebase investigation: multiple subagents explore different file groups, each returning only summaries
- Migration: subagents investigating each unit return only a 3-way classification: "ready / needs manual work / unclear"
- Incident response: separate subagents investigate logs, metrics, and traces, returning only correlated anomaly summaries
⚠ Pitfalls
If distillation quality is poor, critical information can be lost. Even with "return only conclusions" instructions, judging what qualifies as a conclusion depends on model capability. Over-distillation can strip away context the parent agent needs for correct decisions. The granularity of distillation must be tuned to the task.
#
HarnessEngineering# #
AIAgent#
Show more