Sam Hosseini

Platform delivery leader | ex-CTO, ex-Amazon TPM | AI agents, token costs, and shipping real work | Building the AI Situation Room 🇺🇸

[Sam Hosseini @thesam]

The State of AI and LLMs: A Bragging Rights Battle with a Twist on Education
In the ever-expanding universe of artificial intelligence, specifically with large language models (LLMs), we're seeing a spectacle that feels like a digital parent-teacher conference. Each day, tech companies are eager to show off their latest AI "child's" accomplishments, much like parents might compare notes on their kids' progress at school. But here, we're not just talking about traditional schooling; we're drawing parallels to public vs. private education in this AI evolution.

The "Look What My AI Can Do" Show
Imagine if AI development was like schooling:

Company A, akin to a prestigious private school, boasts, "Our LLM can now articulate complex sentences with the finesse of a seasoned poet. It's like our student just aced the literary honors class."
Company B, more like a well-funded public school, counters with, "Well, ours not only talks but also understands context better than ever, practically rolling over in conversational agility. It's as if our student just won the debate championship."
Company C chimes in, representing perhaps an innovative charter school, "Our model has just 'graduated' with flying colors, mastering multiple languages with native-like proficiency, much like a student excelling in a multilingual immersion program."
And then there's Company D, maybe a tech startup like a new, experimental school, proudly stating, "Our AI can now play the piano, or at least, help you compose music with a click. It's like our student just got first chair in the orchestra."

This constant braggadocio isn't just for show; it's a race to claim the top spot in AI innovation, where each new feature is a step towards more human-like, versatile AI interaction.

Behind the Brag: What's Really Going On?
Innovation at Lightning Speed: The pace at which new LLMs are developed mirrors the speed at which educational approaches evolve, especially in private or specialized schools where resources are abundant.
Diverse Applications: Just as education systems vary from public to private, offering different focuses, each LLM promises to revolutionize another sector. Whether it's enhancing medical diagnostics, teaching languages, or improving customer service, the parallels to educational specialization are clear.
The Ethical and Privacy Conundrum: Similar to concerns over how public and private schools handle student data, AI development raises questions about privacy, ethical AI behavior, and data usage. The transparency and governance in AI are akin to debates on educational policy and student rights.
The User Experience: Ultimately, the quality of education or AI interaction is judged by outcomes. Does an LLM that can "play the piano" provide a real educational value, or is it just a showcase of capability?

The Future of AI Bragging
As we look forward, this trend of showcasing AI milestones will likely continue, much like schools celebrating their students' achievements. The competition among AI developers is not just about innovation but also about shaping the future of technology, education, and society.

In this analogy, public AI development might be seen as providing broader access to technology, much like public education aims for universal access, while private AI initiatives might focus on specialized, cutting-edge applications, akin to private schools offering tailored education experiences.

So, while we marvel at each new "Ah" and every roll-over, let's also consider how these advancements reflect and influence our educational systems, our society, and the future we're forging with AI.

[Sam Hosseini @thesam]

Reading the whole book is becoming a quiet edge.
AI summaries hand you the conclusion. The book gives you the reasoning, and that's what lets you ask the better question in the meeting.
Pick one book this month and finish it cover to cover. Audiobooks count.

[Sam Hosseini @thesam]

@trq212 The bitter lesson for teams: general methods beat clever hacks, but someone still has to define done and check it. That part stays human for a while.

[Sam Hosseini @thesam]

@ryancarson A year from now the gap won't be model access. It'll be which teams rebuilt planning and review for agent-speed work and which still run sprints around it.

[Sam Hosseini @thesam]

Stop making these 5 mistakes with AI pilots:
1) No baseline, so no proof
2) Picking the flashiest use case, not the most painful
3) No owner after the demo
4) Security review saved for week 10
5) Calling it a win because people liked it
Make them and your pilot stays slideware.

[Sam Hosseini @thesam]

@shl The new "don't check Slack at dinner" is "don't check what your agents merged at dinner."

[Sam Hosseini @thesam]

Where your agent runs code is now a compliance question.
Google just let Managed Agents on the Gemini API execute in your own sandboxes while planning stays in their cloud.
For regulated teams, that's the gap between "cool demo" and "security will sign off."

[Sam Hosseini @thesam]

@ammaar Coding agents on a phone makes the shift obvious. Nobody types code on that screen. The whole job becomes reading diffs and saying yes or no.

[Sam Hosseini @thesam]

If you're a generalist who worries you never specialized, read Range by David Epstein. It's about why broad experience often wins in messy, fast-changing fields. You'll notice it when the hardest problem crosses three teams. Encouraging, and it changes how you see your career.

[Sam Hosseini @thesam]

@danshipper Solo output is easy to scale. The hard part is a team where everyone ships like five people and nobody knows what anyone else changed. Coordination is the bottleneck.

[Sam Hosseini @thesam]

Honest question for engineering managers: what's the one thing you still won't let an AI agent do on your team, even if it could?
Mine: approve its own change to production.

[Sam Hosseini @thesam]

@emollick Coding moved fast because it already had the feedback loop: tests, CI, code review. Other professions will move at the speed they can verify the work.

[Sam Hosseini @thesam]

@MarcosRGjr Love the stop-only rule. A pre-approved halt is way easier to trust than an auto-fix, and it buys the human time to actually show up.

[Sam Hosseini @thesam]

Dylan Patel says always-on agents went from zero to 45% of SemiAnalysis's AI spend in under ten months.
That's not a tool anymore. That's a service.

Services get an owner, a budget, a dashboard, and an on-call. Agents should too.

[Sam Hosseini @thesam]

Read this before your next performance review cycle.
England's university regulator warns students can finish assessments in minutes with AI, often undetected.
Work has the same problem: polished docs are cheap now. Judge how people reason live, not what they paste.

[Sam Hosseini @thesam]

@omarsar0 Evals are the spec the business never wrote down. Teams that skip them can't tell if a new model is better or just different.

[Sam Hosseini @thesam]

If you're a burned-out builder, read Essentialism by Greg McKeown. It's about doing fewer things on purpose instead of saying yes to everything. You'll feel it the next time you accept a meeting you know you shouldn't. Short, calm, and it changes how you guard your week.

[Sam Hosseini @thesam]

@poteto Congrats. 11,000 PRs a month at ~0.5% reverts is the stat I'd frame. Turning each correction into a check instead of another prompt line is how agents scale.

[Sam Hosseini @thesam]

Hot take: some of the best founding teams of the next decade will be an ops veteran, a seasoned engineer, and someone who's sold into the industry for 15 years.
Each knows a different third of the problem.
Have one of those thirds? Reply and tell me which.

[Sam Hosseini @thesam]

@stretchcloud The path overrides are the real feature. A risk score is a guess. "Never auto-merge src/auth" is a policy, and regulated teams only sign off on policies.

[Sam Hosseini @thesam]

@Codie_Sanchez Pay matters most, but tie pay to one metric and people game the metric. Clear goals plus fair pay beats both speeches and pure incentive plans.