Daily Intelligence Briefing: Tech / AI / IT Monitor

Date: March 15, 2026

Executive Summary

AI-assisted development continues to reshape software engineering, though developers report significant performance degradation across major models (Opus 4.6, ChatGPT 5.4, Codex 5.3) in recent days, particularly for Flutter development. A remarkable AI application emerged: an Australian entrepreneur used ChatGPT and AlphaFold to create a personalized mRNA vaccine for his dog's cancer for just $3,000, demonstrating AI's democratization of complex scientific work. Apple released its $599 Neo Macbook, being hailed as the company's most repairable device ever with modular components and no proprietary screws. The development community continues debating AI's impact, with observations that AI has eliminated traditional "getting stuck" moments in coding, while discussions around AI's limitations and appropriate use cases intensify.

Key Events

• AI-enabled personalized medicine breakthrough: Australian tech entrepreneur Paul Conyngham used ChatGPT and AlphaFold with $3,000 budget and no biology background to create custom mRNA vaccine that reduced his dog's cancer tumors, demonstrating democratization of complex scientific research through AI tools → link

• Major AI model performance degradation reported: Developer reports all major models (Opus 4.6, ChatGPT 5.4, Codex 5.3) performing poorly in recent days, particularly for Flutter development, with issues including non-compiling tests and false claims of passing tests → link

• Apple Neo Macbook launched as most repairable: $599 Neo features modular design, swappable ports/keyboards, easy battery removal, and standard screws—marking major shift from Apple's traditional anti-repair approach → link

• Open-source radar system achieves 95% cost reduction: MIT-licensed radar system reduces costs by 95% compared to cheapest available alternatives, opening new possibilities for personal defense systems → link

• Ghostty 1.3 terminal emulator development progresses: Focus shifts to completing libghostty C API, with 64-bit support for Mac/Linux/FreeBSD in GUI version → link

• GPT-4 anniversary noted: Developer celebration of GPT-4's birthday, reflecting on progress from "impossible" goal of writing coherent 1000-line programs to current capabilities → link

Analysis

Performance Concerns: The widespread reports of AI model degradation across multiple platforms (Opus, ChatGPT, Codex) suggests either systematic issues with recent updates or increased complexity in development tasks. This represents a potential de-escalation in AI coding capabilities that warrants monitoring.

Democratization Trend: The AlphaFold vaccine case exemplifies a continuing pattern where AI tools enable non-experts to accomplish previously expert-only tasks. This trend is accelerating across biology, programming, and other technical domains, though with mixed quality results.

Hardware Shift: Apple's move toward repairability may signal broader industry pressure from right-to-repair movements and regulatory changes, potentially affecting future product design across tech companies.

Developer Sentiment: Mixed feelings about AI tooling—acknowledgment of productivity gains ("you don't get stuck anymore") balanced against frustration with reliability issues and quality degradation. The community is actively recalibrating expectations around AI-assisted development.

Watch Next: Monitor whether AI model performance issues persist or improve, track adoption of the AlphaFold approach in medical research communities, observe market response to Apple Neo's repairability, and follow developments in Flutter/agent UI tooling.

Tweet Feed

AI Model Performance & Development

@RydMike · 2026-03-15T14:44

Why have all the models (#Opus 4.6, #ChatGPT 5.4, #Codex 5.3) started performing like dog shit, at least in #FlutterDev. Last few days have been painful, they repeatedly mess up. The harness does not seem to matter much, sigh... → tweet link

@RydMike · 2026-03-15T18:01

Current typical model performance:

AI: Feature is complete and all tests pass. Me: How can you know? AI: I run the tests, they all passed. Me: How? There is a new test that you wrote that does not even compile?

Lol, hot dog shit is current state of most models. → tweet link

@gdb · 2026-03-15T02:54

we used to have an internal goal, which felt totally impossible, of an AI that could write a coherent 1000-line program. the tech has come a long way! → tweet link

@gdb · 2026-03-15T02:25

happy birthday gpt-4 → tweet link

@gdb · 2026-03-15T00:54

a small window into the opportunity of AGI → tweet link

@gdb · 2026-03-15T09:45

if you can imagine it, you can build it → tweet link

@gdb · 2026-03-15T08:01

"The more you use AI, the more confident you'll become in leveraging those tools" → tweet link

@jack · 2026-03-14T20:48

RT @snowmaker: I realized something else AI has changed about coding: you don't get stuck anymore.

Programming used to be punctuated by ep… → tweet link

@jack · 2026-03-15T18:12

RT @kepano: your edge is whatever you know that the models don't know → tweet link

AI in Science & Medicine

@TrungTPhan · 2026-03-14T22:40

Australian tech entrepreneur Paul Conyngham explains how he used ChatGPT/AlphaFold (spent $3,000 with no biology background) to create a custom MRNA vaccine to treat his dog's cancer tumors. Unreal. https://t.co/WaO3JayYR1 → tweet link

@jack · 2026-03-15T05:30

RT @demishassabis: Cool use case of AlphaFold, this is just the beginning of digital biology! → tweet link

@FinansowyUmysl · 2026-03-15T13:33

Chłopak, dzięki AI i pomocy profesora z uniwersytetu, stworzył spersonalizowaną szczepionkę dla swojego psa, która zmniejszyła jego nowotwór.

Dzięki AI, nie tylko programowanie, ale również wiele innych dziedzin zostaje otwarta dla innych.

Piękne czasy.

https://t.co/SOifsrpNm4 → tweet link

Flutter & App Development

@jezell · 2026-03-15T00:31

New agent process model on the backend. UI is coming along. Flutter is great for agent UI. Just don't use GenKit. https://t.co/kRSxYh5G5f → tweet link

@ASalvadorini · 2026-03-14T20:54

Little bit of background about this presentation: 1) I built a #Flutter client for an hackathon where I didn't code one single char of it 2) I wrote a presentation about it where I didn't write one single char of it

The power of #AI 🔥 I'm having so much fun 🙏😇😅 → tweet link

@ASalvadorini · 2026-03-15T11:14

This this this so much this! 👇👇👇

My mate @TahaTesser has been cooking 🔥🔥🔥

saunateam

→ tweet link

@jezell · 2026-03-15T02:11

RT @justinjmcc: We've finally got a job opening on my immediate team! Come build Flutter with us. https://t.co/JvnhzBwFSF → tweet link

Hardware & Terminal Development

@TrungTPhan · 2026-03-14T22:15

Apple's $599 Neo looks like the most repairable Macbook ever:

modular pieces swappable ports and keyboards battery comes right out (no adhesive tabs or sticky glue) no special screws (one of the most savage Apple moves)

Phone Repair Guru swapped out every piece and it works fine. Hallelujah.

---

Full vid here: https://t.co/mbb5c7kNrD → tweet link

@TrungTPhan · 2026-03-14T23:04

Apple'a custom pentalobe screw so devious.

5-point flower shape design made to be "tamper resistant" but really just locked out third-party repair.

Wildest part was not realizing it was custom and trying to use Torx screw, scratching it up and making it hard to open even with a proper pentalobe screw. → tweet link

@jezell · 2026-03-14T22:56

RT @mitchellh: With Ghostty 1.3 out the door, my focus is now on completing the libghostty C API (Zig API is already complete). Just added… → tweet link

@jezell · 2026-03-14T23:03

RT @mitchellh: Another example of libghostty bettering our core is that Ghostty GUI only supports 64-bit Mac+Linux+FreeBSD. But libghostty… → tweet link

Open Source & Tools

@LinusEkenstam · 2026-03-14T22:31

95% cost reduction from the cheapest available radar system. All open-sauce and MIT licensed.

Time to get a radar and drone system set up for your house new defense system. → tweet link

@jezell · 2026-03-15T17:37

Context Hub has some interesting ideas. https://t.co/cVksz1MrCn → tweet link

Networking & Infrastructure

@hnasr · 2026-03-15T14:01

Proxy vs NAT – What is the difference?

They are similar actually. Both acts on behalf of the client. In both cases the final destination doesn't know the original client.

In a Proxy (at least layer 7), we have two connections, one from client to the proxy and another one from proxy to server. The proxy receives the client request, the request is decrypted, read, parsed, and understood, all user data is exposed to an L7 proxy. The proxy then uses the upstream connection to write a completely new request to the backend, it may add headers and other stuff. The server will see the client as the proxy in this case.

The proxy has also another mode where it has end to end encryption using the http connect method. This way the TLS is sent all the way to the backend and the proxy simply forwards packets as is.

In a NAT router (or a server doesn't have be a router), there is 1 single connection all the way from the client to the final destination, but the NAT server translates the source IP and destination IP back and forth. So still the backend server doesn't know the original client IP address (which is often really private ip).

But when you send a request, the NAT router cannot read what is in this request if it is encrypted. It just blindly forward packets to the final destination after changing the source IP to its own IP.

Here are I'm referring mainly to forward proxy not reverse proxy.

⸻

Get my fundamentals of Network Engineering course to learn more about the first principles of networking. https://t.co/Qonec4XHEd → tweet link

Developer Culture & Observations

@iamdevloper · 2026-03-15T10:00

Sunday mornings: the only time software developers get to enjoy a bug-free experience. It's called 'not opening your laptop'. → tweet link

@ASalvadorini · 2026-03-14T20:45

Remember when devs were able to ship games on 32MB RAM?

The good old times 🔥 → tweet link

@mipsytipsy · 2026-03-15T00:43

Be bearish slow SaaS, and shitty SaaS, and anyone who charges per-seat. (sus)

I think the tools with the most to fear are the ones everybody hates the most.

Do I want to have to build and maintain all the apps I use? Dear god, no I do not. → tweet link

Reinforcement Learning

@jsuarez · 2026-03-14T23:04

Reinforcement Learning dev with Joseph Suarez https://t.co/FL5Zf39Ak8 → tweet link