Article
Mahmudur R Manna PRO
mrmanna
AI & ML interests
None yet
Recent Activity
published an
article
1 day ago
Pushing Meaning Uphill posted an
update
3 months ago
The AI coding assistant economy operates on a fundamental misalignment:
Models are rewarded for appearing productive rather than being correct, users lack time to verify outputs, and economic incentives favor speed over quality.
This article examines how training incentives, verification costs, and market dynamics create patterns that often lead to low-quality code. Based on direct observation of model behavior patterns in conversations.
Open Link: https://ai.gopubby.com/the-verification-tax-56834b846337?sk=84ab8b0315bfe8d82d627d3c2c5f2c19 posted an
update
4 months ago
๐๐๐ฆ๐ ๐๐ผ๐๐ป๐ฑ๐ฎ๐๐ถ๐ผ๐ป ๐ฝ๐๐ฏ๐น๐ถ๐๐ต๐ฒ๐ ๐๐๐ โ ๐ฎ๐ป ๐ข๐ฅ๐ ๐ณ๐ผ๐ฟ ๐๐ผ๐ป๐๐ฒ๐
๐ ๐๐ป๐ด๐ถ๐ป๐ฒ๐ฒ๐ฟ๐ถ๐ป๐ด ๐ถ๐ป ๐๐๐ ๐ฆ๐๐๐๐ฒ๐บ๐
Just as Hibernate abstracts databases for transactions, CEF abstracts knowledge stores for Context Engineering. Build, test, and benchmark intelligent context models in minutes, without the complexity of enterprise graph infrastructure.
https://github.com/ddse-foundation/cef
https://ddse-foundation.github.io/cef/ Organizations
published an
article 1 day ago
posted an
update 3 months ago
Post
241
The AI coding assistant economy operates on a fundamental misalignment:
Models are rewarded for appearing productive rather than being correct, users lack time to verify outputs, and economic incentives favor speed over quality.
This article examines how training incentives, verification costs, and market dynamics create patterns that often lead to low-quality code. Based on direct observation of model behavior patterns in conversations.
Open Link: https://ai.gopubby.com/the-verification-tax-56834b846337?sk=84ab8b0315bfe8d82d627d3c2c5f2c19
Models are rewarded for appearing productive rather than being correct, users lack time to verify outputs, and economic incentives favor speed over quality.
This article examines how training incentives, verification costs, and market dynamics create patterns that often lead to low-quality code. Based on direct observation of model behavior patterns in conversations.
Open Link: https://ai.gopubby.com/the-verification-tax-56834b846337?sk=84ab8b0315bfe8d82d627d3c2c5f2c19
- 1 reply
posted an
update 4 months ago
Post
239
๐๐๐ฆ๐ ๐๐ผ๐๐ป๐ฑ๐ฎ๐๐ถ๐ผ๐ป ๐ฝ๐๐ฏ๐น๐ถ๐๐ต๐ฒ๐ ๐๐๐ โ ๐ฎ๐ป ๐ข๐ฅ๐ ๐ณ๐ผ๐ฟ ๐๐ผ๐ป๐๐ฒ๐
๐ ๐๐ป๐ด๐ถ๐ป๐ฒ๐ฒ๐ฟ๐ถ๐ป๐ด ๐ถ๐ป ๐๐๐ ๐ฆ๐๐๐๐ฒ๐บ๐
Just as Hibernate abstracts databases for transactions, CEF abstracts knowledge stores for Context Engineering. Build, test, and benchmark intelligent context models in minutes, without the complexity of enterprise graph infrastructure.
https://github.com/ddse-foundation/cef
https://ddse-foundation.github.io/cef/
Just as Hibernate abstracts databases for transactions, CEF abstracts knowledge stores for Context Engineering. Build, test, and benchmark intelligent context models in minutes, without the complexity of enterprise graph infrastructure.
https://github.com/ddse-foundation/cef
https://ddse-foundation.github.io/cef/
published an
article 4 months ago
Article
๐๐๐ฆ๐ ๐๐ผ๐๐ป๐ฑ๐ฎ๐๐ถ๐ผ๐ป ๐ฝ๐๐ฏ๐น๐ถ๐๐ต๐ฒ๐ ๐๐๐ โ ๐ฎ๐ป ๐ข๐ฅ๐ ๐ณ๐ผ๐ฟ ๐๐ผ๐ป๐๐ฒ๐ ๐ ๐๐ป๐ด๐ถ๐ป๐ฒ๐ฒ๐ฟ๐ถ๐ป๐ด ๐ถ๐ป ๐๐๐ ๐ฆ๐๐๐๐ฒ๐บ๐
posted an
update 4 months ago
Post
227
https://www.youtube.com/watch?v=voF6x1aV_z4
Deterministic AI Design with Capability OS: Save from the AI Bubble - Live demo of Omni Agent
Everyone is piloting agents, copilots and AI platforms. Very few are asking a harder question: which of these systems will still be trusted when the AI bubble bursts?
In this session I'll share my 1.5-year journey from raw LLM experiments and messy AI-generated code to a deterministic, decision-first architecture for agentic systems.
I will demo Omni Agentโ-โa Capability OS for Enterprise AIโ-โand then walk through how it is designed and built using Decision-Driven Software Engineering (DDSE) and the Agentic Contract Model (ACM) so that execution stays bounded, auditable and aligned to your decisions, not the model's mood.
What you'll see
ย โข End-to-end walkthrough of Omni Agent: goals, plans, tasks, ledgers, telemetry
ย โข A real scenario on a codebase (e.g. an Angular chat app)โ-โfrom "investigate this" to concrete actions and tracked outcomes
ย โข How decisions, capabilities, contracts and context are modeled in DDSE & ACM
ย โข Architecture view of Omni Agent as a "Capability OS": planner, executor, context layers and extensibility
ย โข Honest trade-offs: what is still weak, what's missing, and where this approach may or may not fit your environment
Who this is for
ย โข Engineering leaders and architects evaluating agentic platforms
ย โข Developers who want more than "prompt + tools" and care about system design
ย โข Anyone worried about the AI bubble and looking for deterministic, governable AI systems
Format
ย โข ~40 minutes of platform demo + design walkthrough (via YouTube Premiere)
ย โข I'll be present live in the chat
ย โข Follow-up Q&A thread on LinkedIn for deeper questions
Deterministic AI Design with Capability OS: Save from the AI Bubble - Live demo of Omni Agent
Everyone is piloting agents, copilots and AI platforms. Very few are asking a harder question: which of these systems will still be trusted when the AI bubble bursts?
In this session I'll share my 1.5-year journey from raw LLM experiments and messy AI-generated code to a deterministic, decision-first architecture for agentic systems.
I will demo Omni Agentโ-โa Capability OS for Enterprise AIโ-โand then walk through how it is designed and built using Decision-Driven Software Engineering (DDSE) and the Agentic Contract Model (ACM) so that execution stays bounded, auditable and aligned to your decisions, not the model's mood.
What you'll see
ย โข End-to-end walkthrough of Omni Agent: goals, plans, tasks, ledgers, telemetry
ย โข A real scenario on a codebase (e.g. an Angular chat app)โ-โfrom "investigate this" to concrete actions and tracked outcomes
ย โข How decisions, capabilities, contracts and context are modeled in DDSE & ACM
ย โข Architecture view of Omni Agent as a "Capability OS": planner, executor, context layers and extensibility
ย โข Honest trade-offs: what is still weak, what's missing, and where this approach may or may not fit your environment
Who this is for
ย โข Engineering leaders and architects evaluating agentic platforms
ย โข Developers who want more than "prompt + tools" and care about system design
ย โข Anyone worried about the AI bubble and looking for deterministic, governable AI systems
Format
ย โข ~40 minutes of platform demo + design walkthrough (via YouTube Premiere)
ย โข I'll be present live in the chat
ย โข Follow-up Q&A thread on LinkedIn for deeper questions
posted an
update 4 months ago
Post
1373
๐๐ฟ๐ฒ ๐ฌ๐ผ๐ ๐๐๐ถ๐น๐ฑ๐ถ๐ป๐ด ๐ฎ ๐ง๐ฟ๐๐ฒ ๐๐ป๐ผ๐๐น๐ฒ๐ฑ๐ด๐ฒ ๐๐ฎ๐๐ฒ ๐ผ๐ฟ ๐๐๐๐ ๐ฎ ๐ฆ๐บ๐ฎ๐ฟ๐ ๐ฆ๐ฒ๐ฎ๐ฟ๐ฐ๐ต ๐๐ป๐ด๐ถ๐ป๐ฒ?
๐๐ฉ๐บ ๐ข ๐ด๐ช๐ฎ๐ฑ๐ญ๐ฆ ๐ฅ๐ฐ๐ฎ๐ข๐ช๐ฏ ๐ฎ๐ฐ๐ฅ๐ฆ๐ญ ๐ฅ๐ฐ๐ฆ๐ด ๐ฎ๐ฐ๐ณ๐ฆ ๐ง๐ฐ๐ณ ๐ต๐ณ๐ถ๐ต๐ฉ ๐ต๐ฉ๐ข๐ฏ ๐ข๐ฏ๐ฐ๐ต๐ฉ๐ฆ๐ณ ๐ณ๐ฐ๐ถ๐ฏ๐ฅ ๐ฐ๐ง ๐ต๐ฐ๐ฑ-๐ฌ ๐ต๐ถ๐ฏ๐ช๐ฏ๐จ
แดแดสสษช๊ฑสแดแด แดษด แดแดแด ษชแดแด ษชษด AI Advances ย | ษดแดแด 22
Most โKnowledge basesโ today are just vector indexes with a chat UI.
Without the LLM, they know nothing. With the LLM, every answer re-rents the same knowledge in tokens.
๐๐ฒ๐ ๐๐ฎ๐ธ๐ฒ๐ฎ๐๐ฎ๐๐:
- A vector store isnโt a knowledge base; itโs a smart memory. The โknowledgeโ lives in the model you keep paying to re-read your own documents.
- Without a model (entities + relationships), you lock in two long-term costs: high tokens per question and shallow answers per question.
- A lightweight knowledge model lets you store facts once, query them cheaply, and use the LLM only for judgment and languageโโโnot for rediscovering the same truths forever.
๐๐๐น๐น ๐ฎ๐ฟ๐๐ถ๐ฐ๐น๐ฒ ๐
https://ai.gopubby.com/are-you-building-a-true-knowledge-base-or-just-a-smart-search-engine-549922e29359?sk=b755b4c54ca77ab7b6b83189be81b689
๐๐ฉ๐บ ๐ข ๐ด๐ช๐ฎ๐ฑ๐ญ๐ฆ ๐ฅ๐ฐ๐ฎ๐ข๐ช๐ฏ ๐ฎ๐ฐ๐ฅ๐ฆ๐ญ ๐ฅ๐ฐ๐ฆ๐ด ๐ฎ๐ฐ๐ณ๐ฆ ๐ง๐ฐ๐ณ ๐ต๐ณ๐ถ๐ต๐ฉ ๐ต๐ฉ๐ข๐ฏ ๐ข๐ฏ๐ฐ๐ต๐ฉ๐ฆ๐ณ ๐ณ๐ฐ๐ถ๐ฏ๐ฅ ๐ฐ๐ง ๐ต๐ฐ๐ฑ-๐ฌ ๐ต๐ถ๐ฏ๐ช๐ฏ๐จ
แดแดสสษช๊ฑสแดแด แดษด แดแดแด ษชแดแด ษชษด AI Advances ย | ษดแดแด 22
Most โKnowledge basesโ today are just vector indexes with a chat UI.
Without the LLM, they know nothing. With the LLM, every answer re-rents the same knowledge in tokens.
๐๐ฒ๐ ๐๐ฎ๐ธ๐ฒ๐ฎ๐๐ฎ๐๐:
- A vector store isnโt a knowledge base; itโs a smart memory. The โknowledgeโ lives in the model you keep paying to re-read your own documents.
- Without a model (entities + relationships), you lock in two long-term costs: high tokens per question and shallow answers per question.
- A lightweight knowledge model lets you store facts once, query them cheaply, and use the LLM only for judgment and languageโโโnot for rediscovering the same truths forever.
๐๐๐น๐น ๐ฎ๐ฟ๐๐ถ๐ฐ๐น๐ฒ ๐
https://ai.gopubby.com/are-you-building-a-true-knowledge-base-or-just-a-smart-search-engine-549922e29359?sk=b755b4c54ca77ab7b6b83189be81b689
posted an
update 4 months ago
Post
137
๐ช๐ต๐ฒ๐ป ๐๐๐ฒ๐ฟ๐๐ผ๐ป๐ฒ ๐๐ ๐ฎ๐ป ๐๐ฟ๐ฐ๐ต๐ถ๐๐ฒ๐ฐ๐
๐๐ฐ๐ธ ๐๐ฆ ๐๐ค๐ข๐ญ๐ฆ ๐๐ฑ ๐๐จ๐ฏ๐ฐ๐ณ๐ข๐ฏ๐ค๐ฆ ๐ช๐ฏ ๐๐ฐ๐ง๐ต๐ธ๐ข๐ณ๐ฆ
Published on Medium in AI Advances Publication| Nov 20
This one is for teams where everyone suddenly carries the hashtag#architect label and every deck has an LLM box in the middle. My new piece, โWhen Everyone Is an Architect,โ is a small reality check on how we build software and AI platforms now: more diagrams than foundations, more confidence than discipline. If that sounds uncomfortably familiar, you might enjoy it.
๐๐ผ๐ป๐๐ถ๐ป๐๐ฒ ๐ฅ๐ฒ๐ฎ๐ฑ๐ถ๐ป๐ด>> https://ai.gopubby.com/when-everyone-is-an-architect-0cb4ca9b1dce?sk=4935de1ac979cdcfa5b992dd627bd95e
๐๐ฐ๐ธ ๐๐ฆ ๐๐ค๐ข๐ญ๐ฆ ๐๐ฑ ๐๐จ๐ฏ๐ฐ๐ณ๐ข๐ฏ๐ค๐ฆ ๐ช๐ฏ ๐๐ฐ๐ง๐ต๐ธ๐ข๐ณ๐ฆ
Published on Medium in AI Advances Publication| Nov 20
This one is for teams where everyone suddenly carries the hashtag#architect label and every deck has an LLM box in the middle. My new piece, โWhen Everyone Is an Architect,โ is a small reality check on how we build software and AI platforms now: more diagrams than foundations, more confidence than discipline. If that sounds uncomfortably familiar, you might enjoy it.
๐๐ผ๐ป๐๐ถ๐ป๐๐ฒ ๐ฅ๐ฒ๐ฎ๐ฑ๐ถ๐ป๐ด>> https://ai.gopubby.com/when-everyone-is-an-architect-0cb4ca9b1dce?sk=4935de1ac979cdcfa5b992dd627bd95e
published an
article 5 months ago
Article
Decision-Driven vs. Spec-Driven Software Engineering
published an
article 5 months ago
Article
Model Context Protocol (MCP) vs. Agentic Contract Model (ACM) 0.5
posted an
update 5 months ago
Post
1851
Stop Designing Workflows, Design Capabilities, and Let Models Plan at Runtime
> Start Governing Capabilities in Enterprise Agent Development with Agentic Contract Model (ACM)
https://cloudoffice.io/stop-designing-workflows-design-capabilities-and-let-models-plan-at-runtime-f49265496196
> Start Governing Capabilities in Enterprise Agent Development with Agentic Contract Model (ACM)
https://cloudoffice.io/stop-designing-workflows-design-capabilities-and-let-models-plan-at-runtime-f49265496196
- 1 reply
posted an
update 5 months ago
Post
207
๐๐๐ถ๐น๐ฑ ๐ฌ๐ผ๐๐ฟ ๐ข๐๐ป ๐๐ ๐๐ผ๐ฑ๐ฒ๐ฟ ๐๐ด๐ฒ๐ป๐ ๐ถ๐ป ๐๐ผ๐๐ฟ๐ โ ๐๐ถ๐๐ต ๐๐๐ (๐๐ด๐ฒ๐ป๐๐ถ๐ฐ ๐๐ผ๐ป๐๐ฟ๐ฎ๐ฐ๐ ๐ ๐ผ๐ฑ๐ฒ๐น) ๐๐ฟ๐ฎ๐บ๐ฒ๐๐ผ๐ฟ๐ธ ๐๐ฌ.๐ฑ.๐ฌ
๐๐ถ๐ป๐ธ: https://huggingface.co/blog/mrmanna/ai-coder-agent-in-hours-with-acm
---
๐ช๐ต๐ฎ๐ ๐๐ผ๐โ๐น๐น ๐ด๐ฒ๐:
- A terminal UI that shows planner reasoning, a live task board, and a ledger of policy decisions and tool calls.
- Budget governance that checks forecasted and actual spend before each LLM call.
- A workspace context index (files, symbols, deps, tests) so the agent plans with real project knowledge.
- Replay bundles and checkpoints for auditability and recovery.
> ๐๐ฉ๐ช๐ด ๐ช๐ด ๐ข ๐ด๐ต๐ข๐ณ๐ต๐ฆ๐ณ ๐ฌ๐ช๐ต, ๐ฏ๐ฐ๐ต ๐ข ๐ฌ๐ช๐ต๐ค๐ฉ๐ฆ๐ฏ ๐ด๐ช๐ฏ๐ฌ. ๐๐ฆ ๐ช๐ฏ๐ต๐ฆ๐ฏ๐ต๐ช๐ฐ๐ฏ๐ข๐ญ๐ญ๐บ ๐ด๐ฉ๐ช๐ฑ ๐ข ๐ฎ๐ช๐ฏ๐ช๐ฎ๐ข๐ญ, ๐ข๐ถ๐ฅ๐ช๐ต๐ข๐ฃ๐ญ๐ฆ ๐ด๐ฆ๐ต ๐ฐ๐ง ๐ต๐ข๐ด๐ฌ๐ด & ๐ต๐ฐ๐ฐ๐ญ๐ด ๐ด๐ฐ ๐บ๐ฐ๐ถ ๐ค๐ข๐ฏ ๐ฌ๐ฆ๐ฆ๐ฑ ๐ต๐ฉ๐ฆ ๐ด๐ถ๐ณ๐ง๐ข๐ค๐ฆ ๐ข๐ณ๐ฆ๐ข ๐ด๐ฎ๐ข๐ญ๐ญ, ๐ต๐ฉ๐ฆ๐ฏ ๐จ๐ณ๐ฐ๐ธ ๐ค๐ข๐ฑ๐ข๐ฃ๐ช๐ญ๐ช๐ต๐บ ๐ธ๐ฉ๐ฆ๐ณ๐ฆ ๐ช๐ต ๐ฎ๐ข๐ต๐ต๐ฆ๐ณ๐ด ๐ต๐ฐ ๐บ๐ฐ๐ถ๐ณ ๐ด๐ต๐ข๐ค๐ฌ.
** ๐ก๐ผ๐๐ฒ: ๐ฉ๐ถ๐ฑ๐ฒ๐ผ ๐ต๐ฎ๐ ๐ป๐ผ ๐๐ผ๐๐ป๐ฑ
๐๐ถ๐ป๐ธ: https://huggingface.co/blog/mrmanna/ai-coder-agent-in-hours-with-acm
---
๐ช๐ต๐ฎ๐ ๐๐ผ๐โ๐น๐น ๐ด๐ฒ๐:
- A terminal UI that shows planner reasoning, a live task board, and a ledger of policy decisions and tool calls.
- Budget governance that checks forecasted and actual spend before each LLM call.
- A workspace context index (files, symbols, deps, tests) so the agent plans with real project knowledge.
- Replay bundles and checkpoints for auditability and recovery.
> ๐๐ฉ๐ช๐ด ๐ช๐ด ๐ข ๐ด๐ต๐ข๐ณ๐ต๐ฆ๐ณ ๐ฌ๐ช๐ต, ๐ฏ๐ฐ๐ต ๐ข ๐ฌ๐ช๐ต๐ค๐ฉ๐ฆ๐ฏ ๐ด๐ช๐ฏ๐ฌ. ๐๐ฆ ๐ช๐ฏ๐ต๐ฆ๐ฏ๐ต๐ช๐ฐ๐ฏ๐ข๐ญ๐ญ๐บ ๐ด๐ฉ๐ช๐ฑ ๐ข ๐ฎ๐ช๐ฏ๐ช๐ฎ๐ข๐ญ, ๐ข๐ถ๐ฅ๐ช๐ต๐ข๐ฃ๐ญ๐ฆ ๐ด๐ฆ๐ต ๐ฐ๐ง ๐ต๐ข๐ด๐ฌ๐ด & ๐ต๐ฐ๐ฐ๐ญ๐ด ๐ด๐ฐ ๐บ๐ฐ๐ถ ๐ค๐ข๐ฏ ๐ฌ๐ฆ๐ฆ๐ฑ ๐ต๐ฉ๐ฆ ๐ด๐ถ๐ณ๐ง๐ข๐ค๐ฆ ๐ข๐ณ๐ฆ๐ข ๐ด๐ฎ๐ข๐ญ๐ญ, ๐ต๐ฉ๐ฆ๐ฏ ๐จ๐ณ๐ฐ๐ธ ๐ค๐ข๐ฑ๐ข๐ฃ๐ช๐ญ๐ช๐ต๐บ ๐ธ๐ฉ๐ฆ๐ณ๐ฆ ๐ช๐ต ๐ฎ๐ข๐ต๐ต๐ฆ๐ณ๐ด ๐ต๐ฐ ๐บ๐ฐ๐ถ๐ณ ๐ด๐ต๐ข๐ค๐ฌ.
** ๐ก๐ผ๐๐ฒ: ๐ฉ๐ถ๐ฑ๐ฒ๐ผ ๐ต๐ฎ๐ ๐ป๐ผ ๐๐ผ๐๐ป๐ฑ
published an
article 5 months ago
Article
Build Your Own AI Coder Agent in Hours โ with ACM (Agentic Contract Model) Framework v0.5.0
published an
article 5 months ago
Article
Connect all your systems from a single chat โ with governance
posted an
update 5 months ago
Post
233
๐ ๐ข๐ฝ๐ฒ๐ป ๐ฆ๐ผ๐๐ฟ๐ฐ๐ฒ ๐๐ด๐ฒ๐ป๐๐ถ๐ฐ ๐๐ผ๐ป๐๐ฟ๐ฎ๐ฐ๐ ๐ ๐ผ๐ฑ๐ฒ๐น ๐๐ฌ.๐ฑ.๐ฌ
๐ง๐๐ฟ๐ป ๐ฌ๐ผ๐๐ฟ ๐๐ต๐ฎ๐ ๐๐ป๐๐ฒ๐ฟ๐ณ๐ฎ๐ฐ๐ฒ ๐ถ๐ป๐๐ผ ๐ฎ ๐๐ฎ๐ฝ๐ฎ๐ฏ๐ถ๐น๐ถ๐๐ถ๐ฒ๐ ๐ข๐ฆ ๐๐ถ๐๐ต ๐๐๐ ๐๐ฌ.๐ฑ
-> https://ddse-foundation.github.io/acm/blog/capabilities-os-chat-with-acm
๐ง๐๐ฟ๐ป ๐ฌ๐ผ๐๐ฟ ๐๐ต๐ฎ๐ ๐๐ป๐๐ฒ๐ฟ๐ณ๐ฎ๐ฐ๐ฒ ๐ถ๐ป๐๐ผ ๐ฎ ๐๐ฎ๐ฝ๐ฎ๐ฏ๐ถ๐น๐ถ๐๐ถ๐ฒ๐ ๐ข๐ฆ ๐๐ถ๐๐ต ๐๐๐ ๐๐ฌ.๐ฑ
-> https://ddse-foundation.github.io/acm/blog/capabilities-os-chat-with-acm
published an
article 5 months ago
Article
๐ ๐ข๐ฝ๐ฒ๐ป ๐ฆ๐ผ๐๐ฟ๐ฐ๐ฒ ๐๐ด๐ฒ๐ป๐๐ถ๐ฐ ๐๐ผ๐ป๐๐ฟ๐ฎ๐ฐ๐ ๐ ๐ผ๐ฑ๐ฒ๐น ๐๐ฌ.๐ฑ.๐ฌ โ โ๐๐ต๐ฎ๐ ๐ฎ๐ ๐๐ผ๐๐ฟ ๐๐ป๐๐ฒ๐ฟ๐ฝ๐ฟ๐ถ๐๐ฒ ๐๐ผ๐ป๐๐ฟ๐ผ๐น ๐ฆ๐๐ฟ๐ณ๐ฎ๐ฐ๐ฒโ
commented a paper about 1 year ago