Sando George @sandogee

Futurist | Software Engineer | Strategic Advisor | Military Officer Opinions are my own. github.com/sandogeorge The City of Gold Joined December 2010

Tweets

948
Followers

95
Following

647
Likes

265

Marcin Paprzycki @marcinpaprzycki

3 months ago

ENERGIES (MDPI) told me that I should promote my work ... They published it... so, here is the promotion ;-) Title: Dynamic Knowledge Management in an Agent-Based Extended Green Cloud Simulator Link: mdpi.com/1996-1073/17/4…

1 2 2 59 0

View Details

Sando George @sandogee

3 months ago

I think about this everyday. The coding models are standing on the shoulders of giants.

Sam Altman @sama

3 months ago

I have so much gratitude to people who wrote extremely complex software character-by-character. It already feels difficult to remember how much effort it really took. Thank you for getting us to this point.

5K 2K 36K 5.7M 3K

0 0 0 6 0

View Details

Sando George @sandogee

3 months ago

The current wave of vibe-coded AI systems resembles the spreadsheet era in how it decentralizes capability and embeds critical business logic outside formal engineering controls, but with a more opaque and potentially more hazardous failure surface.

0 0 0 10 0

View Details

Sando George @sandogee

4 months ago

Reading: Reuters / AP (Feb 17, 2026) — fast overview + wider context. reuters.com apnews.com

0 0 0 20 0

View Details

Sando George @sandogee

4 months ago

Reading: Ofcom update (Feb 3, 2026) — what online-safety enforcement can/can’t do for AI chatbots. ofcom.org.uk/online-safety/…

1 0 0 30 0

View Details

Sando George @sandogee

4 months ago

Grok isn’t being judged on “AI potential.” It’s being judged on the distribution loop: personal data → generation → virality → harm. EU/UK regulators are testing whether “safety by design” is real or just a blog post. Where should liability sit?

1 0 0 20 0

View Details

Justin Skycak @justinskycak

4 months ago

Do it scared. Do it tired. Do it doubting. Do it imperfect. Do it alone. Do it anyway.

Noctara @rumilyrics

4 months ago

You literally just have to get really good at continuing.

93 6K 25K 465K 2K

7 160 1K 29K 265

View Details

Sando George @sandogee

4 months ago

@marcinpaprzycki That's a deep philosophical question 🤣

0 0 0 9 0

View Details

Sando George @sandogee

4 months ago

Is 100% test coverage still unrealistic/suboptimal when it's agents designing and writing tests?

1 0 0 29 0

View Details

Sando George @sandogee

4 months ago

@marcinpaprzycki Full agreement that formal verification is the only TRUE 100% test. I'm mainly referring to common unit/integration testing here though. I decided to try and was able to get one of my code bases to 100% in about 12hrs using GPT-Codex-5.3. I had done about 53% manually.

1 0 0 37 0

View Details

Sando George @sandogee

4 months ago

It just hit me that we're probably not authoring GitHub project readmes for humans anymore.

0 0 0 20 0

View Details

Sando George @sandogee

4 months ago

GPT-5.3-Codex. The vibe is strong with this one. This is the way.

0 0 0 63 0

View Details

Sando George @sandogee

4 months ago

@Jayyanginspires Writing forces a level of accountability that is often ignored when speaking.

0 0 0 5 0

View Details

Sando George @sandogee

4 months ago

Remember when boilerplate generation tools made you feel good as a coder?

0 0 0 30 0

View Details

Sando George @sandogee

4 months ago

So will the mythical 10x engineer now be replaced by a solid 1x engineer doing agentic coding?

0 0 0 31 0

View Details

Sando George @sandogee

4 months ago

For years I've thought of the idea of cloning myself to get more work done. Now we have fairly capable agents that are easy to spin up and direct. I'm happy.

0 0 0 27 0

View Details

Sando George @sandogee

4 months ago

Reading: arxiv.org/pdf/2601.09760 — shows how tool vs memory conflicts can quietly break reliability; provenance makes the mismatch visible.

0 0 0 21 0

View Details

Sando George @sandogee

4 months ago

Reading: arxiv.org/pdf/2601.20727 — argues for “LLM audit trails” (tamper-evident provenance) as a practical accountability mechanism.

1 0 0 35 1

View Details

Sando George @sandogee

4 months ago

AI will draft you a plan that sounds like it survived a staff brief. But if you can’t trace where each claim came from, you’re outsourcing judgment—not typing. Habit: demand provenance (“source / assumption / confidence”) before you act.