Ranjan Roy from Margins is back for our weekly discussion of the latest tech news. We cover: 1) OpenAI tells us how people use ChatGPT 2) Practical guidance is the top use of ChatGPT 3) Is generative AI actually a threat to search given the use cases? 4) OpenAI has a very broad definition of 'doing' or agent work 5) The hidden impact of AI 'decision support' in the economy 6) People trust AI bots massively - is that bad? 7) ChatGPT's massive growth 8) Anthropic shares Claude's economic uses 9) Automation is surpassing augmentation for AI in work 10) Will Meta's AI glasses hit? 11) Can Jimmy Kimmel build an audience off-ABC? 12) Will the next Jimmy Kimmel be a youtube/rpodcaster?
---
Enjoying Big Technology Podcast? Please rate us five stars ⭐⭐⭐⭐⭐ in your podcast app of choice.
Episode overview:
April Long spent two years fighting reality. The co-founder and CEO of "Afro-Asia Cross-border payment infrastructure" startup Pyxis was so determined to serve Africa's small merchants - the "bottom of the pyramid" she'd read about in Harvard Business Review - that she nearly bankrupted her fintech ignoring the bulk traders actually driving Africa-China trade.
In conversation with Andile Masuku, Long delivers uncomfortable truths about impact theatre versus impact reality. Her journey from receiving President Xi Jinping in Tanzania at 23 to finally accepting who actually moves goods between Africa and China at 35 offers a masterclass in entrepreneurial humility.
Key insights:
-On impact delusions: "I used to defend, I was like, 'No, no, no, no, no. It's that you don't get to this market.'" Long admits she lived in a bubble, desperately wanting to believe SMEs were ready for direct China trade. The truth? "90% of African trade is still happening in a more traditional way" - through the aggregators she'd dismissed as insufficiently mission-driven.
- On the cost of stubbornness: Despite zero demand after six months embedded in Nairobi's wholesale markets, Long refused to pivot. "I was quite stubborn. I was like, no, we have to work with SMEs." The result: burning 90% of her time on unprofitable small traders whilst the 10% spent on bulk traders kept her company alive.
- On acceptance as strategy: "The future is not here yet. And we need to build the future by serving who is there currently." Long's breakthrough came from accepting that Chinese trading companies scaling from $0 to IPO in a decade were the real infrastructure of Africa-China trade - not the romantic vision of empowered individual merchants.
- On being un-fundable forcing clarity: Without millions to burn on market education, Long had to face reality faster than her funded competitors. "I'm grateful I didn't have money to burn, or else I could have burned myself."
Notable moments:
1. The marketplace wake-up call: Walking through Nairobi's famous Gikomba market as a Chinese woman, traders shouted "China, China, what are you selling?" They wanted products, not payment rails. Long built the wrong solution for the right market.
2. The Eric Simanis paradox: The same Harvard Business Review article that inspired her Africa move warned against oversimplifying "bottom of pyramid" markets. Long spent years learning what she'd initially misread.
3. The three Aprils: Long describes fragmenting into Chinese April, Western April, and African April - "these narratives are so vastly different" that keeping them separate became exhausting. Building Pyxis became about reconciling these selves.
The aggregator revelation:
Long's former Standard Chartered clients - the Chinese trading companies she'd tried to convince to take loans in 2015 - transformed from traders to manufacturers to near-IPO giants in under a decade. They were the real story of Africa-China trade, moving containers whilst she chased individual merchants moving parcels.
"These Chinese trading companies making impacts in Africa, making products super affordable... because of the storytelling, they are not recognised." Her role shifted from trying to bypass them to helping them operate more efficiently.
The present tense:
Long's current focus on settlement infrastructure for bulk traders isn't the sexy SME empowerment story she'd imagined. But with a 12-person team across four countries and actual revenue, she's building what the market needs today whilst preparing for the SME future she still believes will come.
Image credit: Pxyis
Ryan welcomes Sebastian Gierlinger, VP of Engineering at Storyblok, to talk about how headless content management systems (CMS) fit into an increasingly componentized software landscape. They run through the differences between headless and traditional CMS systems (and databases), prototyping and security concerns, and how a team building distributed systems can get that precious velocity by decoupling their content from its rendering.
Episode notes:
Storyblok provides a headless CMS they say is made for humans but built for the AI-driven era.
Want to learn more about CMS design? Check out other pieces we’ve done with CMS providers Drupal and Builder.io.
Today on Talk Python: What really happens when your data work outgrows your laptop. Matthew Rocklin, creator of Dask and cofounder of Coiled, and Nat Tabris a staff software engineer at Coiled join me to unpack the messy truth of cloud-scale Python. During the episode we actually spin up a 1,000 core cluster from a notebook, twice! We also discuss picking between pandas and Polars, when GPUs help, and how to avoid surprise bills. Real lessons, real tradeoffs, shared by people who have built this stuff. Stick around.
Aaron Levie is the CEO of Box. Levie joins Big Technology to discuss the reports that a vast majority of businesses are not getting a return on their AI investments. Levie shares his takeaways from the reports, gives a rebuttal, and discusses the reality on the ground. Stay tuned for the second half where we separate hype from reality in the AI agent conversation. Tune in for a wide-ranging, post-Boxworks deep dive on where AI is heading in the coming years.
---
Enjoying Big Technology Podcast? Please rate us five stars ⭐⭐⭐⭐⭐ in your podcast app of choice.
Jens Neuse grew up in Germany, originally planning to be a carpenter. In his 2nd year as an apprentice, he was in a motorcycle wreck that thrust him into a process of surgery and healing. Eventually, he decided he wouldn't be doing carpentry, and got into sysadmin work. Once he got bored with this, he moved into startups, learned how to code, and starting digging into programming, API's and eventually - GraphQL federation. Outside of tech, he is married with 3 young kids. He loves to sit ski on the mountain - which is the coolest carbon fiber chair on a ski, where you steer with your knees and hips.
After chasing building a better Apollo, Jens and his team ran into a point where their prior product and company was doomed to go under. When they accepted this fact, they started to think about what people actually wanted - and started to dig into the federation of GraphQL.
Ryan chats with Karen Ng, EVP of Product at HubSpot, to chat about Model Context Protocol (MCP) and how they implemented it for their server for their CRM product. They chat the emergence of this as the standard for agentic interactions, the challenges of implementing the server and integrating it with their ecosystem, and how agentic AI has affected work at Hubspot.
Episode notes:
Hubspot is a customer-relationship management (CRM) platform that aims to help businesses grow.
MCP is an open-source protocol for connecting AI agents to external systems, originally developers at Anthropic.
Join us on YouTube at pythonbytes.fm/live to be part of the audience. Usually Monday at 10am PT. Older video versions available there too.
Finally, if you want an artisanal, hand-crafted digest of every week of the show notes in email form? Add your name and email to our friends of the show list, we'll never share it.
A judge lets Google keep paying Mozilla to make Google the default search engine but only if those deals aren’t exclusive.
More than 85% of Mozilla’s revenue comes from Google search payments.
The ruling forbids Google from making exclusive contracts for Search, Chrome, Google Assistant, or Gemini, and forces data sharing and search syndication so rivals get a fighting chance.
Note that just saying you require 3.9+ doesn’t tell the user that you’ve actually tested stuff on 3.14. I like to keep Trove Classifiers around for this reason.
Also, License classifier is deprecated, and if you include it, it shows up in two places, in Meta, and in the Classifiers section. Probably good to only have one place. So I’m going to be removing it from classifiers for my projects.
One problem, classifier text has to be an exact match to something in the classifier list, so we usually recommend copy/pasting from that list.
But no longer! Just use troml!
It just fills it in for you (if you run troml suggest --fix). How totally awesome is that!
I tried it on pytest-check, and it was mostly right. It suggested me adding 3.15, which I haven’t tested yet, so I’m not ready to add that just yet. :)
pqrs is a command line tool for inspecting Parquet files
This is a replacement for the parquet-tools utility written in Rust
Built using the Rust implementation of Parquet and Arrow
pqrs roughly means "parquet-tools in rust"
Why Parquet?
Size
A 200 MB CSV will usually shrink to somewhere between about 20-100 MB as Parquet depending on the data and compression. Loading a Parquet file is typically several times faster than parsing CSV, often 2x-10x faster for a full-file load and much faster when you only read some columns.
Speed
Full-file load into pandas: Parquet with pyarrow/fastparquet is usually 2x–10x faster than reading CSV with pandas because CSV parsing is CPU intensive (text tokenizing, dtype inference).
Example: if read_csv is 10 seconds, read_parquet might be ~1–5 seconds depending on CPU and codec.
Column subset: Parquet is much faster if you only need some columns — often 5x–50x faster because it reads only those column chunks.
Predicate pushdown & row groups: When using dataset APIs (pyarrow.dataset) you can push filters to skip row groups, reducing I/O dramatically for selective queries.
Memory usage: Parquet avoids temporary string buffers and repeated parsing, so peak memory and temporary allocations are often lower.
Brian #4: Testing for Python 3.14
Python 3.14 is just around the corner, with a final release scheduled for October.
Ranjan Roy from Margins is back for our weekly discussion of the latest tech news. We cover: 1) Apple's impressive new iPhone Pro models 2) Who is the iPhone Air for? 3) Has the phone reached its ultimate form factor 4) Is generative AI threatening to upend the smartphone market 5) Meta's new smartglasses are coming 6) Nepal's Gen Z overthrows the government and picks a new leader on Discord 7) OpenAI growth stats after GPT-5 launch 8) Oracle and OpenAI's new $300 billion deal 9) Flirting with ChatGPT 10) The AI companionship use case is real 11) Does San Francisco have 996 work culture?
---
Enjoying Big Technology Podcast? Please rate us five stars ⭐⭐⭐⭐⭐ in your podcast app of choice.
Want a discount for Big Technology on Substack + Discord? Here’s 25% off for the first year: https://www.bigtechnology.com/subscribe?coupon=0843016b
Questions? Feedback? Write to: bigtechnologypodcast@gmail.com
Ryan welcomes Geraint North, AI and developer platforms fellow at Arm, to dive into the impact of GenAI on chip design, Arm’s approach to designing flexible CPU architectures, and the challenges of optimizing large language models at the chip level for edge devices.
Episode notes:
Arm is a global compute platform that allows the world’s leading technology companies to innovate and deliver AI experiences.
Arm just announced their Lumex CSS Platform, which provides a complete compute subsystem platform for mobile and desktop providers to enable efficient AI workloads.
