The Sunday Brew | Issue #2, Aug‘26 | Free
Welcome to The Sunday Brew, weekly 1-2-3 newsletter by The Percolator. Every Sunday we drop in your inbox 1 story in a picture, 2 concepts, ideas or frameworks to expand your horizons and 3 news from the week, to keep you updated.
If you are not a paid subscriber, here is what you missed last week:
ONE STORY IN A PICTURE
TWO IDEAS, FRAMEWORKS OR CONCEPTS
This week we bring to you two Concepts: Theory of Constraints & Antifragility
The Theory of Constraints dictates that the maximum throughput of any operational network is strictly governed by its most restrictive component.
When professionals design processing pipelines, they must recognise that the overall output is not determined by the sum of individual efficiencies but by the capacity of the slowest node. Any physical or computational sequence possesses a specific limitation where the accumulation of inputs exceeds the processing rate. Accelerating non-constraining components creates excess inventory and local overproduction, which degrades the overall stability of the architecture. Systemic improvement requires isolating this exact restriction and directing resources to expand its specific capacity.
A common structural error occurs when teams attempt to optimise every module simultaneously. Because the primary restriction dictates the absolute boundary of the final output, engineering effort spent on non-constraining sectors yields zero net increase in total production. Professionals must instead subordinate all other processes to the pace of the primary bottleneck. Upstream operations should only supply inputs at the precise rate the restriction can process them to prevent structural blockages. Once the initial limitation is resolved, the constraint naturally shifts to a different component within the network.
Managing this dynamic demands a rigorous cycle of addressing sequential limitations as they emerge. By treating operational capacity as a quantifiable flow controlled by the most restricted conduit, teams allocate capital with high precision. This disciplined methodology ensures that resource expenditure directly translates to measurable increases in total output. Maintaining this focus prevents the misallocation of effort and allows for the sustained expansion of functional capability.
🚀
Antifragility defines a structural property where an architecture actively expands its functional capacity when subjected to volatility and external pressure.
Unlike resilient designs that merely withstand shocks without degrading, these frameworks require disorder to trigger their own improvement. When exposed to variable stressors, the system utilises these environmental inputs to reorganise its internal components into a more capable configuration. This property relies on allowing localised failures to occur safely, which then forces the broader structure to overcompensate and build excess capacity against future variations.
In complex engineering, attempting to eliminate all variance often creates rigid structures that fail catastrophically under unpredicted loads. Professionals designing adaptive networks must instead incorporate intentional stress mechanisms to maintain operational fitness. By exposing a system to continuous and low-level volatility, engineers provide the necessary data for the architecture to calibrate its own defences. The damage sustained at the micro level serves as an instructional signal, directing resources to fortify the exact points of vulnerability before a systemic failure can occur.
Constructing these dynamic frameworks demands a shift from passive defence to active adaptation. Leaders must design decentralised networks where individual modules can fail independently without compromising the overarching structure. This distributed approach ensures that the destruction of a single node generates protective knowledge that immediately benefits the surviving components. By treating volatility as a necessary catalyst for growth rather than a threat to be minimised, professionals can build systems that continually optimise their own performance in direct proportion to the environmental pressure they encounter.
🚀
An Earnest Appeal
The Percolator is built like a magazine, but without the team or organisation. It takes research, writing, editing, and a steady rhythm to publish longform essays twice a week. Paid subscriptions make this work possible, and sustainable.
If you’ve found value here, I’d love for you to consider upgrading. Paid subscribers unlock every essay, can comment and join the chat, and a chance to feature their own notes and stories.
👉 Please upgrade to paid and help keep the brew strong. It costs less than two coffees a month, but unlike coffee, the energy lasts longer.
THREE NEWS FROM THE WEEK
SpaceX has outlined an ambitious vision in which autonomous robots establish industrial infrastructure on the Moon before human workers arrive.
The proposal, discussed during the company’s first public earnings call, would involve humanoid or specialized robotic systems constructing factories, maintaining equipment and processing lunar resources in the harsh environment.
Elon Musk acknowledged that the plan currently sounds “totally nuts” but argued that robotic construction on the Moon could eventually become a vital part of a much larger space-based economy. The company has not disclosed a development timeline, budget or confirmed robotic platform. It also remains unclear whether Tesla’s Optimus humanoid robots would be used or whether SpaceX would develop machines specifically designed for lunar conditions.
One concept mentioned was a solar-powered mass accelerator capable of launching materials from the Moon into space without conventional rockets. Such a system would require extensive construction, energy and maintenance infrastructure, potentially making autonomous robots essential to its development.
The lunar proposal forms part of a broader strategy involving artificial intelligence and orbital computing. SpaceX reportedly plans to work with Nvidia on Starmind AI-1, an orbital data-centre satellite using Nvidia’s Rubin GPUs and Vera CPUs. Initial satellites are expected to launch next year, with prototype testing planned for early 2027.
Although the announcements suggest a long-term convergence of robotics, artificial intelligence and space manufacturing, they remain conceptual. SpaceX has yet to provide concrete technical or financial details, leaving the Moon-factory vision closer to a strategic moonshot than an active programme.
➖
ByteDance, the Chinese technology giant behind TikTok, is training an artificial intelligence model with up to 10 trillion parameters, positioning it to compete directly with Anthropic's most advanced Mythos system, according to people familiar with the matter.
The pre-training phase is expected to last three to six months, and if completed at the reported scale, the model would exceed the size of Chinese startup Moonshot AI's Kimi K3, which has 2.8 trillion parameters, by more than threefold. Industry estimates place Anthropic's flagship Mythos 5 at roughly 8 trillion parameters, suggesting ByteDance's ambition places it at or above the current frontier of AI development.
The move follows a directive from ByteDance founder Zhang Yiming instructing staff to halt reliance on distilling rival systems' outputs and instead pursue independent capability building, a shift that comes amid accusations from Anthropic that several Chinese firms used Claude to train competing models.
ByteDance's Seed research team, established in early 2023, now employs about 2,000 people globally and is led by Wu Yonghui, a former Google and DeepMind researcher who joined as research head in February 2025. Unlike many Chinese competitors that favour open-source approaches, ByteDance keeps most models closed-source, though its Doubao chatbot remains China's most popular AI service with 324 million monthly active users.
The development highlights intensifying competition between Chinese and American technology companies at the cutting edge of artificial intelligence, with recent models from Moonshot and Alibaba receiving favourable reviews in comparative benchmarks. ByteDance did not respond to requests for comment.
➖
Meta is facing fresh criticism after an ABC News Verify investigation found that its Facebook Content Monetisation programme paid advertising revenue to creators linked to far-right and anti-vaccine movements.
The invitation-only scheme, which disbursed nearly US$3 billion to an estimated 16.2 million accounts in 2025, rewards creators based on the performance of eligible public posts, a model critics say financially incentivises content that provokes strong reactions.
Among those identified were Hugo Lennon, a far-right agitator removed by Victoria Police in Melbourne for directing racist abuse at Indian Prime Minister Narendra Modi, and The Noticer, a website described as promoting white supremacist and neo-Nazi ideologies. Also included were pages associated with the anti-immigration group March for Australia and Monica Smit, founder of anti-lockdown group Reignite Democracy Australia, whose page featured vaccine misinformation while marketing “radiation protection” bracelets.
Meta declined to comment on specific accounts but said creators must comply with its policies or risk losing access to ad revenue. The company draws a line between offensive speech and content that can cause real-world harm, stating it does not seek to “police offensiveness”. However, accountability advocates argue the programme effectively places Meta in a commercial relationship with participating creators, raising concerns about consistency in enforcing rules that bar misleading medical information and restrict monetisation on controversial social issues.
The report underscores a broader policy dilemma for platforms that rely on engagement-driven algorithms: whether paying creators for reach and reactions indirectly rewards material that undermines public discourse, even when it does not breach explicit content rules.
The Sunday Brew by The Percolator brings to you curated news on tech, business & entrepreneurship, from across the internet to give your week a perfect start.
Share your thoughts and opinions on the topics covered in this newsletter by leaving a comment and joining the conversation.
Comments
Nothing yet. Say the first thing.
Sign in to join the conversation.