"There are very few companies on the planet capable of building and deploying these complex AI systems. If you like to build, Meta is the place for you.” – Steve Mills, Mechanical Engineer Ever wondered what it’s like for your work to impact billions? Steve Mills, mechanical engineer, and Dharmesh Jani (DJ), open systems hardware lead, share how innovating at Meta empowers them to solve complex challenges across the AI stack and take their skills to new heights.

🛠️Some engineering patterns show up at every scale — from a $2 shim under a machine bed to platforms serving billions. 📐Shims have been the quiet heroes throughout my career: from metal shims correcting hardware tolerances on the manufacturing floor, to software shims resolving Windows DLL/path conflicts when systems don’t quite line up. 💪Last week, I used my trusty dial gauge + shims to level my 3D printer bed. This week, I read how Meta applied the same idea to modernize WebRTC (behind Messenger, Instagram video, and Meta Quest) after years of drift from upstream open source. 😄 💡Their move was elegant: a shim layer — a thin proxy API that let legacy + latest WebRTC implementations coexist, enabling A/B testing across 50+ use cases instead of a risky big‑bang rewrite. 🎯 Results (per Meta): - Up to 10% CPU improvement - Up to 3% better crash rates - 87% less dual‑stack overhead vs duplicating higher layers 👏 As a product leader and systems thinker, I can’t help but admire solutions that let teams move fast while staying safe. 🤔 What’s the most elegant “shim” you’ve seen — in code, hardware, or even process? (Full Meta article linked in the comments ⬇️)

🤖 Meta projects that 75% of code will soon be generated by AI. But one engineering team just paid a 94,000 dollar price tag for a single incident because their agents lacked business context. That is the reality of the double transformation hitting our industry. We are moving from simple code completion to full workflow orchestration. When agents start deploying code, the senior engineer's job description changes overnight. You stop being a coder and start being a super-operator. Your value isn't in how fast you type, but in how deeply you understand the business invariants that the AI simply cannot see. At YozmaTech, we have seen this shift coming. It is why we prioritize seniority—70% of our 200 engineers are mid or senior level. With an average tenure of 4+ years, our teams have the historical context needed to scrutinize what the agents produce. You can automate the syntax, but you cannot automate the tribal knowledge. We are training our leads to design these agent workflows rather than fear them. The goal is not to replace the engineer, but to remove the friction between the code and the customer. 🧠 👉🏻 How much of your weekly workflow is actually being handled by autonomous agents right now? And more importantly—do you trust them with the production environment? Curious to hear your experience. #AIEngineering #SoftwareArchitecture #EngineeringCulture #YozmaTech

Mark Zuckerberg lost the AI war at a 2013 dinner party. Why the smartest founder in AI turned down Meta’s biggest check for less money. DeepMind’s Demis Hassabis was fielding acquisition offers. Zuckerberg offered a massive, life-changing payout. Larry Page offered significantly less. So Hassabis ran a psychological test. Over dinner at Zuck's house, he deliberately pivoted the conversation toward VR, AR, and 3D printing. Zuckerberg lit up, treating them all as equally exciting. To Hassabis, that was the ultimate tell. Zuckerberg saw AI as just another shiny feature. Page saw it as the foundation of the future. But here is the hidden math most founders get wrong... Average founders obsess over optimizing for valuation. Generational builders optimize for infrastructure. DeepMind didn’t need Meta’s cash to win. They needed Google’s servers. Page’s winning pitch wasn't a dollar amount; it was a structural advantage.. Don't waste the best years of your life building a company from scratch. Take my compute and solve intelligence tomorrow. Hassabis used Meta's massive cash offer purely as leverage to squeeze Google's M&A team, then walked away to take the lower bid. This has to be on of the most crazy stories i have came across recently

gotta spend the tokens, not waste them, but transform tokens to value. I have been improving on my total token usage for the past 2 years. I am making consistent improvements, what's the upper limit? That's yet to be seen, and as the models improve at long horizon tasks its more pheasable to run more concurrent autonomous agents.

That’s a classic Goodhart’s Law kind of situation: “When a measure becomes a target, it ceases to be a good measure.” Or in other words: give a group of people a metric to improve upon and it will immediately become useless. Once people know they are being judged by a metric, they optimize for the metric itself – not the underlying goal. This happened so many times in the past: - Lines of code → people write more code, not better code - Response time KPIs → quick but low-quality replies Number of tokens used → just run agents in endless loops reviewing and rewriting each other’s work. Without human judgement and interpretation the metrics mean nothing.

Don't know whether any of this is actually true. No means for me to actually find out (nor any incentive to even try). But here's what comes to mind : "Irrational exuberation" "Economy" in Trumpistan seems to be merely about dollars shifting from owner to owner by the gazillions, without any true plus-value ever being created, except perhaps the one that derives from the faulty assumption that machines seem to get ever smarter, when in reality it's only the humans that are getting ever dumber.

If even a fraction of this is true, Meta just encountered a billion dollar incentive problem. The reports are wild. One engineer allegedly burned 281 billion tokens in 30 days. Across the company, the internal leaderboard (called "Claudeonomics") tracked 60 trillion tokens in a month. At public pricing, that is roughly $900 million in compute. Meta’s CTO believes high spend correlates with 10x productivity. NVIDIA’s CEO even suggested offering $250K in AI compute as a signing bonus. But the system is already breaking. Engineers are reportedly gaming the rankings. They are running agents for hours just to climb the "Claudeonomics" board and earn titles like "Token Legend." This is a classic case of the Cobra Effect. During British rule in India, authorities wanted to reduce the cobra population. they offered a cash reward for every dead snake. The result? People started breeding cobras to collect the bounty. When the government realized this and stopped the reward, the breeders released the useless snakes. The cobra problem ended up much worse than when they started. In economics, this is called Goodhart’s Law: When a measure becomes a target, it ceases to be a good measure. By gamifying token usage, Meta has turned "cost" into a "proxy for value." But running a high-frequency API loop is not the same as writing better code. The engineers racing to the top of these leaderboards aren't proving they are irreplaceable. They are proving that their activity is just noise. The Bottom Line: The AI industry needs to implement serious controls immediately. We cannot treat compute as an infinite resource to be "burned" for status. We need to shift the focus from activity to outcomes. If we don't put guardrails on unit economics and actual value created, we are just subsidizing the most expensive ego trip in corporate history. #AI #Engineering #TechStrategy #Economics #Meta

🚀 Meta just made a power move that could reshape the entire AI infrastructure game. They announced FOUR generations of custom AI chips (MTIA 300, 400, 450, 500) rolling out through 2027. Here's why this matters: **The Real Story:** Meta isn't just building chips—they're breaking free from Nvidia's grip. MTIA 400 is already competitive with leading commercial products, and they're cutting costs while doing it. **What This Means for You:** ✅ **Vertical Integration is the New Moat** — Companies that control their own silicon will dominate. If you're in tech, this is the playbook. ✅ **Nvidia's Monopoly is Cracking** — Competition drives innovation. Expect better chips, lower prices, and more options across the industry. ✅ **Custom AI Infrastructure = Competitive Advantage** — The future belongs to companies that optimize their entire stack, not just buy off-the-shelf solutions. **The Bigger Picture:** This is what happens when you have $100B+ to invest. Meta is essentially saying, "We're not waiting for vendors—we're building the future ourselves." The companies that follow this playbook will win. The ones that don't will be left paying premium prices for commodity hardware. What's your take—is custom silicon the future for every major tech company, or just the mega-players? 👇 #AI #TechInfrastructure #Meta #Nvidia #ArtificialIntelligence #FutureOfTech

Meta Platforms Ltd is reportedly pulling top software engineers from across the company into a newly created AI unit on a mandatory basis, with the stated goal of eventually having autonomous agents perform the bulk of the work of building, testing, and shipping its products, and human engineers serving only to monitor them. The development was based on an internal company memo authored by Maher Saba, a vice president in Meta’s Reality Labs division and a longtime associate of Chief Technology Officer Andrew Bosworth, who leads the new Applied AI (AAI) Engineering organization, reported Reuters. According to the report, Saba created AAI last month and initially sought volunteers to join. This week, he told selected employees their transfers are no longer a choice. https://lnkd.in/dAC5KemJ #Meta #MovesFastToward #AworldWhereAIBuildsTheSoftware