Most agent frameworks are LLM-first: the conversation loop is the core, tools are attached to it, rules layer on top, and logging is bolted on at the end for observability. State is persisted as retrievable "memory."
forward deployed engineers were basically what every engineer was in the 90s and 2000s.
normal people who could code, speak to customers, understand the business and ship the thing.
we had to recoin the term because the average dev became a cave goblin who hates meetings, users and sunlight.
happy to see fdes making a comeback
MAI and other SLMs can now be tuned right from every enterprise's Azure cloud subscription/regional deployment with clear data residency agreements, without a vendor contract (AdaptiveML, PremAI, Predibase). That's what we call enterprise lock-in!
If done correctly, this could bring a lot of $$ for Microsoft.
It’s time to move from renting intelligence to truly controlling your AI. Microsoft Frontier Tuning lets you take our models and make them uniquely your own, turning them from capable generalists to completely custom partners.
It starts with reinforcement learning environments (RLEs) that allow our models to learn directly from your workflows. Think of them as training gyms for AI. Here the agent learns your very specific processes, your standards, your way of working. It goes from off-the-shelf to hyper-adapted to exactly what you and your teams need. Those adaptations drive efficiency and performance, and your unique models can keep continually learning in your RLEs. This changes the nature of AI – and it changes the impact.
For example, within Microsoft we use our RLEs combined with our MAI models to climb towards the best agentic use cases for Excel. Our MAI tuned model is on par with GPT-5.4 on public and private benchmarks, while being up to 10X more efficient.
Only you control your agents made with Frontier Tuning. You keep the benefits of your hard-earned know-how, data and institutional knowledge. With us, the RLEs and the models you build in them become your moat.
This is distinct. It’s a new era. An era of AI that you control, on your terms. I think it’ll be a good one. More on the blog: https://t.co/v65eop5aHS
The core move is to invert the stack. Make the append-only event log the source of truth. Everything else — the working graph, the agent's beliefs, and task state — becomes a deterministic projection of that log.
Most agent frameworks are LLM-first: the conversation loop is the core, tools are attached to it, rules layer on top, and logging is bolted on at the end for observability. State is persisted as retrievable "memory."
The core move is to invert the stack. Make the append-only event log the source of truth. Everything else — the working graph, the agent's beliefs, and task state — becomes a deterministic projection of that log.
Last week we hosted one of the greatest demo nights in canadian history during @TOtechweek alongside @OpenAIDevs@buildfutureto had a full house inside of Scotiabank Theatre. A theatre i used to marvel at as a kid.
as a practitioner in the ai space i get to work with a lot of companies at different stages of their journey, and i wanted to give the stage to other builders so you (and i) could see how they actually think
this is a technology with no single right way to build, so sharing how others approach it matters
we lined up some great demo's by
@burkelibbey of @Shopify demo'd a world exclusive first of river (before their own shopify event)
@Rasmic cloned among us and build a simulation with different gpt models. set out to figure out which ones are honest and which ones are liars (gpt 5.3 is con artist)
@dhasandev of @trySquadHQ showcased a software factory, with governance. this kid is destined for greatness.
@trillhause_ showcased a self-healing codebase that was build with codex, with @worklayer@joshgonsalves_ demo'd his codex-powered agent artifact app called foleo
nights like this take a village:
@robjama@miketanx@ForsterCPhoto@alspee@_LindsayCraig
I have one that is authwalled but runs on cloudflare tunnels and cron refreshed daily, dont need an insights layer as id ask my hermes to query the json data.
Yo @xai team, this would be an amazing demo of @grok capability.
Push button, have it read all your bookmarks, organise them, make a report on the most interesting one and your interests over time etc
Everyone is bullish on India until they start their own business and they need 15 papers of bhang bhosda and a current account just to integrate a payment gateway.
building harnesses involve 2 things:
1. feature addition
2. constraint addition
if you figure out what tool/stack does 1 and what thought/primitive does 2, you are golden.
How to Delete 18 Years of Academic Mediocrity and Build a High-Agency Mindset
LETTER FROM A FATHER
Dear Son/Daughter,
For 18 years, you were chasing a gold medal in a system that is fundamentally flawed. The issue is far deeper than NEET paper leaks and CBSE exam glitches.
A Sea of Institutionalized Mediocrity
The Indian education system was originally designed to create clerks for the British Empire. It was meant to reward academic obedience rather than ignite the mind and challenge the status quo.
That system never really changed in independent India.
At the 2005 Golden Jubilee celebrations of AIIMS, Dr. Manmohan Singh famously noted: “Neither institutions nor individuals can survive as islands of excellence in an environment of mediocrity.”
The Foundations of Success
Look at the foundations of the world’s most successful economies. They don’t succeed by accident. They succeed because their education systems are progressive. They are designed for agency, risk-taking, and building for tomorrow.
Our education system, by contrast, is designed to demoralize and defeat the vast majority.
So, my child, here is my counsel to you. If the system will not change, you must change. You must delete 18 years of standardized compliance and build a high-agency mindset.
High Grades, Low Agency
Why is such a large share of India’s youth unemployable in the knowledge economy? Because they possess "Academic Intelligence" but zero "Market Agency."
Our education system is stuck at the very bottom of how humans learn:
a. The School Trap: 90% of your time was spent on memorizing without understanding. "Originality" was penalized as "going out of syllabus."
b. The New Reality: In 2026, Large Language Models (LLMs) perform memorization tasks with 99.9% accuracy at zero marginal cost. The "correct answer" you stressed over for 18 years is now a free API call.
c. The Competition: While you memorize textbooks and learn the tricks of coaching institutes to clear entrance exams, progressive economies focus on the top of the pyramid: experimenting and creating. They train builders; we are still training clerks.
The Psychological Prison
In school, you were conditioned with behavioral traps that are toxic to a modern builder. You must break out of your mental cage.
a. Fear of Failure: In entrance exams like JEE, NEET, or UPSC, a single mistake on an OMR sheet can be the end of road to an “island of excellence.”
Failure in school or college leads to a lifetime of trauma. But in reality, innovation is a series of controlled failures. In the world of science and technology, failure is a badge of honour.
b. Permission-Seeking Culture: While you wait for your teacher’s “instructions” and follow them without questioning, the progressive global education systems encourage High Agency.
What is High Agency: The innate ability to bend reality, to find a way around obstacles, and to build solutions without asking for permission.
c. Degree Obsession: My child, stop obsessing over a Tier-1 college tag as your ticket to success. The world is turning to skill-based hiring. If your teacher didn’t tell you this, let me enlighten you that a piece of paper no longer defines your worth or capabilities.
d. Follower Mindset: Your school taught you to solve “closed problems” (clean data, one right answer.) That’s what made us a nation of great coders, and poor inventors. If you continue to do that, you will be competing with AI tools that cost $20 a month.
The world rewards “open problems” (messy data, infinite answers, uncertainty.) Your value no longer lies in finding the “correct” answer to someone else’s question. Your value lies in identifying a problem, and solving it when outcomes are uncertain.
Survival Guide to Unlearning
a. Shift from “Syllabus” to “Stack”: Stop asking, "What should I learn?" and start asking, "What can I build?" Pick a project that genuinely scares you: an AI app, a micro-SaaS, a community initiative, or a hardware prototype. Do not join a coaching institute. Learn by doing and failing.
b. Practice Strategic Disobedience: In any project, feel free to challenge a standard operating procedure. If something exists simply “because it’s always been done that way,” be sure it is waiting to be disrupted. High agency begins the moment you stop asking for permission to improve things.
c. Replace Rote with First Principles: Elon Musk popularized this. First Principles thinking is the antithesis of everything they taught you at school. Break every problem down to its most fundamental truths (whether it’s physics, economics, or biology) and build your understanding from ground up.
d. Build a Proof of Work (the New CV): Your public footprint is your new degree. Whether it is your GitHub repository, a Substack newsletter, a Twitter community, or a portfolio of real-world projects, publicly document your journey of hurdles.
That creates a permissionless career. You don’t need an HR manager to "clear" you; you need a global market to validate you.
Final Advice: No Pain, No Gain
The system of rote learning and mastering exam skills gave you a strong discipline and an ability to work harder than your global peers. But this system also tried to steal your curiosity.
My child, now let the world be your lab. Get ready to take risks and fail. As Nvidia’s Jensen Huang said:
“Unless you have a tolerance for the pain of failure, you will never experiment. For all of you Stanford students, I wish upon you ample doses of pain and suffering.”
With love, Dad
This is the actual bottleneck. The models are smart enough already. What is missing is the company-specific context locked in senior people heads. Whoever cracks knowledge extraction at the company level unlocks the rest.
As you work on this, please consider using GBrain as your OSS retrieval layer
https://t.co/0F5uDQzPHu
Company Brain will be the last mile problem for Microsoft and its customers. And Microsoft IQ could be that missing piece of the puzzle. Yet to see if OSFI auditors come in the middle with infinite Corrective Action Plans (CAP).
Here it comes.
Microsoft's AI chief just dropped a bombshell: all white-collar work will be automated within 18 months.
Not some jobs. Not most jobs. ALL white-collar work.
This isn't some distant sci-fi prediction - they're talking about next year.
The same company behind ChatGPT's wild growth is now putting a hard deadline on when your spreadsheet-wielding, email-writing, meeting-attending job gets handed over to AI.
Whether he's right or just really bullish, one thing's obvious: the people building this stuff think it's moving way faster than most of us realize.