Even LLMs can run within a browser.

If you are more interested in the history and what has made this possible, check out my Substack post. open.substack.com/pub/soundofd...

If you think of looking at your email, calendar, or even doing a quick thing on Photoshop — all of them are already running in the web without the need to install a piece of software or to do all the heavy lifting on the server.

Some companies stopped trusting browsers and their fast-evolving field, pushing all the logic to the servers, consuming tons of energy and hardware. Some are even able to sell that as a feature called streaming. But in the end, modern browser applications will prevail.

Somehow, we made it through, but I still believe there's software out there running on these old technologies, and they still found a way to package it.

Remember when Flash was turned off, or when ActiveX was disabled? As a consumer, it didn't have much of an impact — well, apart from Flash games disappearing — but for enterprise software, it was a total mess.

Distance to Working Software

Debating strategy with 1,000 people when 10 could ship a prototype is the core failure mode.
Make it cheaper to try than to argue: tiny teams, weekly demos, kill fast.
Optimize lead time to learning.

The gap between AI adopters and everyone else is widening rapidly. Success now hinges on clear specifications, documentation, and thorough testing, as it always has. Coding itself was never the real bottleneck; it was everything else in the development process.

open.substack.com/pub/soundofd...

- Claude Code: CLI powerhouse with exceptional reasoning
- Sourcegraph amp: Innovative "think hard" mode + oracle debugging
- Cursor: Most approachable UI for newcomers
- Kiro: AWS's spec-driven approach yields higher quality code
- GitHub Copilot: Enterprise favorite with tight GitHub integration

Have you faced similar pressures? How do you ensure technical voices are heard?

In my latest Substack post, I explore the crucial importance of trust and clear communication between engineers and management.

Check it out:
👉 open.substack.com/pub/soundofd...

The Therac-25 story still echoes: When our software affects human safety, we carry an immense responsibility. Past lessons must guide future innovation. (5/5)

Today's lesson? As we build AI, self-driving cars, and robotic surgery systems, we can't forget: Code isn't just ones and zeros. It's about human lives. (4/5)

This incident transformed healthcare tech forever. We got mandatory incident reporting, stronger FDA oversight, and better testing protocols. Most importantly? A wake-up call about software's real-world impact. (3/5)

The culprits? Race conditions, poor error handling, and removed hardware safeguards. A perfect storm of preventable issues. (2/5)

5/ Key learning: Career growth means adapting to slower feedback loops. The higher you climb, the more patience and strategic thinking you need.

How do you handle the transition from immediate to long-term feedback?

3/ Team leads shift from code to people. Success isn't measured in commits anymore, but in team growth and delivery. The feedback? Much slower.

4/ At Director/CTO level, you're playing the long game. Strategic decisions, stakeholder management, and company-wide impact. Feedback cycles? Think years.

1/ Engineers thrive on instant feedback - compilers, CI/CD pipelines, and code reviews provide immediate validation. It's satisfying and direct.

2/ Moving to architect/staff engineer roles? Get ready for longer feedback cycles. Your technical decisions might take years to prove right (or wrong).

4/ Key insight: AI tools work best with simpler tech stacks. Going forward, this will be crucial for staying competitive.