Medium

Andrii Tkachuk

Pinned

In

Data Science Collective

by

Andrii Tkachuk

·

17h ago

The Local Harness Is Part of Your SDLC

You don’t need a hundred Skills and an MCP server for everything. What Revolut’s 2,000-engineer rollout actually shows, and what a local…

The Local Harness Is Part of Your SDLC

Pinned

In

Data Science Collective

by

Andrii Tkachuk

·

Oct 2

The Harness Is the Product: How Agents Are Actually Built in 2026

The model brings intelligence. The harness decides whether you can control it, direct it, and trust it in production.

The Harness Is the Product: How Agents Are Actually Built in 2026

Pinned

In

Data Science Collective

by

Andrii Tkachuk

·

Mar 15

Stop Building AI Apps for Every Idea. Start Building MCP Servers.

Why building a full AI app is no longer the smartest default — and why MCP servers are becoming the new capability layer.

Stop Building AI Apps for Every Idea. Start Building MCP Servers.

In

Data Science Collective

by

Andrii Tkachuk

·

Sep 14

Stop Returning JSON. Start Shipping UI with MCP Apps

MCP was already good at giving agents capabilities. MCP Apps finally gives humans a proper interface to those capabilities.

Stop Returning JSON. Start Shipping UI with MCP Apps

In

Data Science Collective

by

Andrii Tkachuk

·

Aug 30

Stop Shipping Individual MCP Servers. Start Shipping Agent Plugins

The next useful abstraction isn’t another tool integration. It’s a portable package of capabilities.

Stop Shipping Individual MCP Servers. Start Shipping Agent Plugins

In

Data Science Collective

by

Andrii Tkachuk

·

Aug 23

Why Browser Execution Is Becoming the Missing Capability in Agent Architecture?

The browser isn’t a new kind of agent — it’s a new execution environment.

Why Browser Execution Is Becoming the Missing Capability in Agent Architecture?

In

Towards AI

by

Andrii Tkachuk

·

Aug 17

Stop Building AI Apps for Every Idea. Start Building MCP Servers — Part #7

MCP servers are becoming platforms, not just tool wrappers.

Stop Building AI Apps for Every Idea. Start Building MCP Servers — Part #7

In

Data Science Collective

by

Andrii Tkachuk

·

Aug 4

AI Agents Should Think in Operations, Not Commands

Your agent doesn’t need to know kubectl, AWS CLI, or gh exists.

AI Agents Should Think in Operations, Not Commands

In

Towards AI

by

Andrii Tkachuk

·

Jul 14

Stop Building AI Apps for Every Idea. Start Building MCP Servers — Part #6

MCP gave agents tools. Code Mode (sandboxes) gives them a runtime.

Stop Building AI Apps for Every Idea. Start Building MCP Servers — Part #6

In

Data Science Collective

by

Andrii Tkachuk

·

Jun 24

AI Security for AI Engineers: What Actually Breaks in Production?

From prompt injection to supply chain: what to actually fix before shipping LLM agents to production.

AI Security for AI Engineers: What Actually Breaks in Production?

Andrii Tkachuk

Andrii Tkachuk

Senior AI Engineer | Applied AI Architect | AI Agents, MCP, RAG, AI Governance, Agent Harness & Сloud-native AI 🧠 https://www.linkedin.com/in/andrii-tkachuk7/

Following

Help

Status

About

Careers

Press

Blog

Store

Privacy

Rules

Terms

Text to speech