Gemini CLI builds research-grounded social copy
Ayaz reports two Gemini CLI builds of a social-post generator: similar architecture, but more platform-specific copy when the agent could consult a research notebook.
blog.codewithahsan.dev
Original prompt
Shared by the source author. See source notes for context.
Build a social media post generator using Node.js, Google Genkit, and the Gemini model.
Reach people exploring AI agents
See placements →What happened
Ayaz reports two Gemini CLI builds of a social-post generator: similar architecture, but more platform-specific copy when the agent could consult a research notebook.
Reported by Muhammad Ahsan AyazHow the agent did it
- Create a notebook with engineering and platform-writing sources.
- Connect Genkit and NotebookLM MCP servers to Gemini CLI.
- Ask the agent to query relevant research before implementing flows.
Where a person stepped in
Ayaz chooses the reference library and test topic, runs the applications and compares the resulting copy. Public posting and engagement gains are not demonstrated.
Muhammad Ahsan Ayaz gave Gemini CLI the same social-post-generator brief twice. Both builds used Node.js, Genkit and the same interface requirements. The second run additionally queried a NotebookLM notebook containing Genkit engineering material and platform-specific copywriting research. The trial topic was a funding announcement used as test input, not a claim that his company had raised that funding.
The connected Genkit MCP server let the agent consult current documentation, start the runtime, execute generated flows and inspect traces. For the research run, it first asked the notebook about LinkedIn hooks, X length, Facebook tone and related conventions, then used those findings while building separate platform flows. Ayaz reports nearly unchanged application architecture but different writing choices, including earlier hooks and more distinct voices across platforms.
The copyable detail is to supply a reference library before generation, then check both structured output and the actual prose. Valid JSON does not establish useful copy. The author provides final flow files and describes qualitative comparisons, but reports no controlled engagement benchmark or proven audience lift. His guide also notes that Genkit MCP was experimental at the time. Readers should choose relevant sources, inspect traces and review each draft before making public marketing claims.
Prompt
Original prompt
Source: Blog · Muhammad Ahsan Ayaz
Frequently asked questions
What differed between the two runs?
The second could query a NotebookLM research collection; both had Genkit documentation access and the same build brief.
What did the author report improving?
Platform-specific writing choices, such as hook placement and tone, rather than a major change in application architecture.
Was better engagement measured?
No. The article describes a qualitative output comparison and does not report a controlled audience-performance benchmark.
More details
Remaining steps
- Have it start the runtime and execute each generated platform flow.
- Compare structured output and read the resulting copy before publication.
About this account
Gemini CLI queries research and current framework documentation, generates platform-specific flows and executes them through MCP before handing off a working generator.
Firsthand account by Muhammad Ahsan Ayaz; original publication date verified against Primary page article:published_time and JSON-LD datePublished 2026-04-26T07:14:20Z. Outcomes are the author’s reports, not independently reproduced. Original prompt excerpt transcribed from the primary account; excerpt only, not the complete prompt. Preserve the source-specific context when adapting it.
Credit: Muhammad Ahsan Ayaz
Source published 26 Apr 2026
Source checked 14 Oct 2026
Tools and topics
Gemini CLI · NotebookLM MCP · Genkit MCP · Node.js · Zod
Research grounding · Social copy · Trace inspection
Editorial completeness: 100/100. Measures documented fields, not agent performance or verified outcomes.