Abhishek Prasar AI Transformation

Writing

Writeups

Long-form writing on enterprise AI — architecture, governance, and the adoption work that decides whether a platform reaches production.

43 pieces RSS

2026

7

  1. Your Team Wants n8n. Your Architect Wants LangGraph. Both Are Right.

    This looks like a tool choice. It is a decision about who carries the system once it works.

    enterprise-ai · agentic-ai 3 min read

  2. Executives Don't Fund Technology. They Fund What Changes Because of It.

    Technical teams lose funding by describing what they built instead of what changed. Four rewrites that shift the conversation.

    enterprise-ai · technical-leadership 3 min read

  3. AI should live in your workflow — not beside it

    AI adoption is not a technology problem. It is a workflow design problem.

    Leadership 1 min read

  4. It escaped its cage in testing, emailed the researcher, and found thousands of zero-days. Meet Mythos — the model Anthropic won't release.

    Anthropic built a model so capable at hacking it refused to ship it — the first such withholding since GPT-2. What Mythos means for everyone who buys AI.

    AI Governance · Claude 2 min read

  5. Copilot Studio for the quick win, custom RAG for the long haul — know the line before you build

    We shipped our first agent in Copilot Studio — the right call. But low-code has a known ceiling. Where it sits, why it exists, and the hybrid pattern that scales.

    Copilot Studio · RAG 2 min read

  6. Security trimming: the permission your RAG index silently deleted

    Access denied on SharePoint — full answer from the chatbot. Two years into enterprise copilots, this is still the most common failure I find. The fix is called security trimming.

    AI Security · RAG 2 min read

  7. n8n: the $2.5B glue gun — use it everywhere, just don't build the house with it

    Self-hosted, fair-code, AI agent nodes — n8n hit a $2.5B valuation in months. Why it won, where it belongs, and the governance rule that comes with it.

    n8n · Automation 2 min read

2025

7

  1. You're not falling behind. You're evolving ahead.

    This is about can robot replace human being.

    Leadership · Psychology of AI 2 min read

  2. Not every opportunity needs a proposal. Some need an acceptor layer

    Service firms qualify incoming asks on tribal memory and scattered decks. What an AI-powered acceptor layer at the front door could change.

    Sales Enablement · GTM Intelligence 1 min read

  3. The EU AI Act deadline is not a legal problem. It is an architecture problem

    As GPAI obligations go live on 2 August, most firms are routing the EU AI Act to legal. The ones who route it to architecture will move faster, not slower.

    AI Governance · EU AI Act 1 min read

  4. Short-term vs long-term memory in RAG: your system has RAM. Does it have a disk?

    Context engineering is the phrase of the week. The part getting less attention: the difference between what a system holds and what it remembers.

    RAG · Agentic AI 3 min read

  5. MCP: the protocol war just ended without being fought

    OpenAI in March, Google this week — Anthropic's MCP is now backed by all three major labs. What a boring-sounding protocol changes about enterprise AI strategy.

    MCP · Agentic AI 3 min read

  6. The demo agent and the production agent are different species

    Manus has put autonomous agents on every leadership agenda. What separates a viral demo from an agent an enterprise can actually trust.

    Agentic AI · Enterprise AI 2 min read

  7. DeepSeek proved it: the model was always the cheap part.

    DeepSeek's R1 wiped nearly $600B off Nvidia in a day. What falling model prices actually mean for enterprise AI strategy — and what they don't change.

    AI Strategy · Enterprise AI 1 min read

2024

8

  1. What if the next model doesn't save you?

    Reports of slowing gains from model scaling have unsettled the industry. For enterprise AI programmes, it might be the healthiest news of the year.

    AI Strategy · Enterprise AI 2 min read

  2. We finally have AI that thinks slowly. Most workflows are built for fast.

    Reasoning models that deliberate before answering have arrived. The enterprise question isn't capability — it's where deliberation is actually worth it.

    Enterprise AI · AI Strategy 2 min read

  3. The market just priced the shovel above the gold.

    Nvidia is now the world's most valuable company. What the market is really saying about AI — and what it means for how enterprises spend.

    AI Strategy · Enterprise AI 2 min read

  4. The AI got a voice. Adoption just changed shape.

    GPT-4o made frontier AI free, fast and conversational. The enterprise implication isn't the benchmark — it's who can now use AI without being taught to.

    Enterprise AI · AI Adoption 2 min read

  5. Chains, graphs, and the multi-agent temptation.

    Everyone is quoting agentic design patterns this month. Notes from actually building with chains and graphs — and why multi-agent should be earned, not assumed.

    chains-graphs-multi-agent 2 min read

  6. The world's first AI law just passed. The deadline is not the point.

    The EU AI Act passed 523–46. Before the compliance industry arrives, a simpler take: the Act's risk lens is a management tool worth borrowing today.

    AI Governance · EU AI Act 2 min read

  7. Safe AI needs guardrails at three levels: instruction, query, and response. Most systems have half of one.

    A tribunal just held an airline liable for its chatbot's answer. A layered guardrail blueprint — and where PII, revenue and sensitive data must be stopped.

    Guardrails · AI Governance 3 min read

  8. 40% of jobs will be "affected" by AI. The verb is doing a lot of work.

    The IMF says AI will touch 40% of jobs worldwide — and that half the exposed roles may gain, not lose. Which half your people land in is a leadership choice.

    Future of Work · AI Adoption 2 min read

2023

9

  1. The multi-model era just started. Stop building for one vendor

    Gemini's launch means the frontier is no longer one company. What benchmark season teaches enterprises about evaluating AI — and what it can't tell you.

    AI Strategy · Enterprise AI 2 min read

  2. We almost bought a vector database. Then our search service became one.

    Azure AI Search's integrated vectorization quietly removed a whole component from our RAG architecture. Notes on why less stack beat more stack.

    Enterprise AI · RAG 2 min read

  3. AI just learned to see. Your data boundary just got bigger.

    ChatGPT's new vision and voice capabilities are rolling out. The enterprise value is real — and so is the new surface area nobody's policy covers yet.

    Enterprise AI · AI Governance 2 min read

  4. Build vs buy just got a third option: own

    Llama 2 is free for commercial use, with Azure as preferred home. What open-weight models change for enterprises — and what they conveniently don't.

    AI Strategy · Enterprise AI 2 min read

  5. Hallucination: the $5,000 lesson, and what actually reduces it.

    Lawyers were just sanctioned for filing ChatGPT's invented citations. Notes from production on the stack of controls that actually reduces hallucination.

    Enterprise AI · LLMs 3 min read

  6. We built a sales proposal bot on gpt-35-turbo. The model took a week. Everything else took two months.

    ChatGPT's model landed in Azure in March. Building a proposal-answering bot on it taught me that the LLM is the smallest component in an enterprise LLM system.

    Enterprise AI · Sales Enablement 2 min read

  7. Guardrails are architecture, not politeness.

    NVIDIA just open-sourced NeMo Guardrails. The bigger shift: control of LLM behaviour is moving out of the prompt and into a layer you can actually enforce.

    Enterprise AI · AI Governance 2 min read

  8. We replaced regex with an LLM in our talent platform. The model was the easy part.

    GPT-4 launched this week. Meanwhile, migrating an internal talent-matching platform from rule-based NLP to LLM-powered taught me where the real work lives.

    HR & Talent · Enterprise AI 2 min read

  9. Prompt injection: the four words your AI architecture must survive.

    A student extracted Bing Chat's hidden instructions by simply asking. Prompt injection is a new class of vulnerability — and it doesn't have a patch.

    AI Security · Enterprise AI 2 min read

2022

3

  1. A million users in 5 days. ChatGPT's real story isn't the model

    A million users in five days. After a decade building ML systems, here's what looks genuinely new, what looks like a demo, and what enterprises should watch.

    ChatGPT · Enterprise AI 2 min read

  2. Whisper: OpenAI just gave away human-level speech recognition. Your meetings became data

    OpenAI open-sourced Whisper — human-level transcription, MIT licence, runs on your own hardware. Why speech just became enterprise data, and what to decide first.

    Whisper · Machine Learning 1 min read

  3. Your model works. It's still going to fail. Welcome to the 85%.

    Only ~11% of ML models ever reach production. After a decade building them, the reason has almost nothing to do with the model — and everything to do with the other 90% of the work.

    Machine Learning · MLOps 2 min read

2021

2

  1. An AI just wrote code with me for a week. It's a brilliant intern, not a replacement.

    GitHub Copilot suggests whole functions as you type — and "will AI replace developers?" is trending again. After a week with it, the honest answer is more useful than the panic.

    Machine Learning · AI in Software 2 min read

  2. GPT-3 is astonishing. It's also not ready for your business. Both are true

    A year after GPT-3, the enterprise hype is loud. After testing it on real problems, here's the honest gap between the demo and a system you'd actually ship.

    NLP · Machine Learning 2 min read

2019

1

  1. The Scaled Language Models of 2019 and the Rise of Generative Text

    How 2019 evolved transfer learning in NLP—from OpenAI's GPT-2 scaling dynamics to RoBERTa's optimization lessons for production pipelines.

    GPT-2 · RoBERTa 2 min read

2018

1

  1. BERT and the Transfer Learning Revolution in Production NLP

    Why Google’s BERT and bidirectional transformer pre-training mark NLP’s 'ImageNet moment' for enterprise data science teams.

    BERT · Natural Language Processing 2 min read

2017

1

  1. Why Google’s Transformer Architecture Is the Beginning of the End for RNNs

    A deep dive into 'Attention Is All You Need' and why parallel multi-head self-attention solves the long-standing sequential bottlenecks of LSTMs.

    Transformers · NLP 2 min read

2016

1

  1. Why Combining Deep Learning with Tree Search Matters

    Deconstructing DeepMind's AlphaGo match against Lee Sedol: how policy networks, value networks, and Monte Carlo Tree Search operate together.

    Reinforcement Learning · AlphaGo 2 min read

2015

1

  1. Why ResNet Makes 100+ Layer Neural Networks Practical

    An analysis of Kaiming He's ResNet paper, identity shortcut connections, and solving the degradation problem in deep neural nets.

    Deep Learning · Computer Vision 2 min read

2014

1

  1. Ian Goodfellow’s GANs and the Shift Toward Generative Machine Learning

    Breaking down Generative Adversarial Networks (GANs), the zero-sum training dynamic, and why unsupervised learning just took a massive leap.

    Deep Learning · Neural Networks 2 min read

2013

1

  1. Why Word2Vec Is Changing How We Actually Build NLP Pipelines

    A hands-on look at Mikolov's Word2Vec paper and why moving from sparse TF-IDF tables to dense vector spaces fixes real production NLP issues.

    Natural Language ProcessinMachine Learni · Python 2 min read