πΊοΈ AI Generated Artifact of This Article
Optimizing AI Workflows: Key AI Working Procedures
To effectively harness artificial intelligence, understanding different AI working procedures is crucial. These models define how humans and AI collaborate for optimal productivity:
- Human First, AI Assist: human is focusing on the high-level tasks and design to instruct AI for high-level decision making or cross-multiple domain tasks.
- AI First, Human Assist in Sync: human design the task and instruct AI to do the task in a synchronous way, and human is focusing on revise and review the AI's work till the task is done.
- AI First, Human Assist in Async: human design the task and instruct AI to do the task in an asynchronous way, AI using
Agentic
approach to do the task, and human is focusing on later review and revise the AI's work till the task is done. These tasks are usually self-contained and not coupled with other domains, so AI can work independently.
Let's take software development as an example:
Human First, AI Assist
- App production, tech-architecture, and market high-level design within the artificial intelligence landscape.
- Core modules, core features, and core functions design.
- Key-decision making about the technology stack, architecture, and modules features, a vital part of AI best practices.
- Complex problem solving and reasoning, such as systematic and systemic reasoning.
- ...
In this mode, Arno try to create a systemic and systematic approach to leverage the power of AI and personal context engineering to do the high-level design and decision making, and then use the AI to assist in the low-level tasks. This approach is central to productivity with AI.
AI First, Human Assist in Sync
- app development, coding, and programming in specific modules
- write PRD, design documents, and technical documents
- code review, PR review, and design review
- software development lifecycle, such as devops, CI/CD, and testing
- ...
In this mode, dedicated AI tools are used to assist in the coding and programming tasks, such as
Cursor
,Github Copilot
, andV0
. These tools are designed to work in sync with human developers, providing real-time assistance and suggestions.
AI First, Human Assist in Async
ALL RUNNING in the background, passively and asynchronously.
- Do market or product research using specialized AI tools.
- Write unit-test and integration-test as part of an AI-driven development cycle.
- Refactor modules, features, and functions, optimizing AI workflows.
- Batch job on data processing or information processing.
- ...
In this mode, AI tools are used to work independently on specific tasks, such as
Google Deep Research
,Manus
andGoogle Jules
which can run in the server-end or background without human intervention. These tools are designed to handle specific tasks and can work asynchronously, allowing for greater efficiency and productivity with AI.
Practical AI Application Scenarios & Tool Recommendations
This section explores various scenarios where AI can be applied, along with recommended AI tools and platforms that fit each AI workflow.
Meta Thoughts
- use Project / workspace of OpenAI ChatGPT or Google's Notebook ML to organize related context info in a project.
- use thinking tool driven for QA and guide AI to dive deeper, with search, deep-reasoning, and
Agentic
tools operations. - use personal context engine to build the context to co-work with AI tools, platforms and services.
- define SOP and use Agentic Tool to batch execute the SOP.
Daily QA
Daily general QA task -> Gemini Pro 2.5 / GPT5, deep understanding and reasoning -> Open AI O3Pro / GPT5
General knowledge and information retrieval.
- public domain knowledge
- time sensitive information
- domain specific knowledge
- experience driven query
- personal context required query
dimension:
- open search based
- private search based
- domain specific pre-trained model based
- personal context based
- direct or reasoning based
AI in Coding & Programming: Tools and Techniques
Augment x Gemini-Cli (Claude 4 / o3 / Gemini Pro 2.5)
Programming and coding tasks.
- vibe-coding & code-generation (codex, augment) for me, VSCode is the HUB for various of agentic tools and AI powered services and MCP
- tabs engineering, quick editing contents as needed
- software-development-lifecycle -> devops agentic tools
- use workspace / project as context QA (in ChatGPT)
AI for Research, Study & Learning: Enhancing Knowledge Discovery
Google Deep Research x Gemini Pro 2.5 x Google Artifact x Google Notebook ML (Knowledge based QA)
or
OpenAI ChatGPT DeepResearch x OpenAI ChatGPT Agent Mode to co-work with content and QA research
- public domain knowledge based
- private domain knowledge based
- professional & profound knowledge based
- short v.s. long form of report (research, study, learning, etc.)
- organize and produce content in one workspace or generate artifacts to share results in need
Arno's SOP for research types of topics.
- create workspace for context gathering and basic QA (in one workspace)
- elaborate topics / problems with QA linked to thinking models, your thoughts, ideas, and experiences (Human First, AI Assist), ask interesting and deep questions and try to use AI to solve and combine your first-thoughts
- DeepResearch topic (shallow, deep x rich and poor context)
- generate basic Artifact (article + webpage)
- DeepResearch + O3-Pro to research topic
- integrate those results together -> Generate Final Artifact (WebPage or Article)
- [optional] add all of those materials and generated content or cited source into a Google Notebook ML for further analysis and insights
AI in Media Generation: Images, Video, and Audio
GPT4o x Midjourney x Google Video Generation
- image generation -> GPT 4o
- video generation
- audio generation
- 3D generation
AI for Complex Problem Solving & Strategic Reasoning
Deep Research x Reasoning models x Context Engine (for context engineering) Agentic Tools for SOP defined Think deep with thinking-tools
- context engineering: KESW x data x info
- complex reasoning
- systematic and systemic context powered
Navigating the AI Landscape: Essential Tools, Platforms & Models
The following legend explains the icons used to categorize these AI technologies:
- π free for now
- π¬ research its features and capabilities
- ππ» recommended, good to have
- β₯οΈ my favorite, must-have
- π valuable but cost
Comparative Guide to AI Tools & LLM Models (2025)
LLM Models Comparison
Latest models recommendations till 2025-06-16
Model | Speed | Price | Best For | Limitations |
---|---|---|---|---|
GPT5 | Medium | ππ» | General tasks | |
Gemini Pro 2.5 | Fast | ππ»β₯οΈ | General tasks, with search context | expensive |
Claude 4 | Medium | ππ»π | Coding |
AI Development Tools for Software Engineering
Tool | Type | Integration | Feature |
---|---|---|---|
Augment | IDE Extension | IDE plugins | πππ»β₯οΈ |
OpenAI Codex | Remote Container Mode | WebApp | ππ» |
Claude Code | Cli Tool | CLI | πππ» |
Google Gemini Cli | Cli Tool | CLI | πππ» |
DeepWiki | Knowledge Base | Docs | π¬ good for open-source project study |
Cursor | IDE | Git, AI models | π limitation grows |
Github Copilot | Code Assistant | IDE plugins | πΏ just so so |
For local IDE VibeCoding, Arno's recommendation is:
Augment > Cursor > Windsurf = Microsoft Github Copilot
AI Tools for General Purpose
Tool | Type | Best For | Integration | Feature |
---|---|---|---|---|
Dia | AI Web Browser | Web browsing and context | API | π¬ browser first assist |
GenSpark | General Purpose | General purpose | API | π¬ almost a copy of ChatGPT without its core AI model |
AI Tools for Research & Knowledge Management
Tool | Features | Best For | Tags |
---|---|---|---|
Google Deep Research | Advanced search, Insights | Deep research | ππ»β₯οΈ |
Google Notebook ML | Knowledge organization | Research synthesis | ππ» |
ChatGPT Research | Research & General Usages | AutoRun tasks | ππ» |
Manus | Research & General Usages | AutoRun tasks | ππ¬ |
Flowith | AI-powered agentic workflow | Parallel tree like ai-powered workflow | π¬ |
AI Creative & Design Tools
Tool | Type | Best For | Tags |
---|---|---|---|
Midjourney | Image Generation | High-quality art | π |
GPT4o | Image Generation | Daily use | π |
Figma Make | AI Design & Impl. | Automated design | π |
AI Deployment & Infrastructure Platforms
Platform | Purpose | Features | Tags |
---|---|---|---|
Vercel | Deployment | Serverless, Edge | ππ» |
OpenRouter | API Gateway | Multiple models | ππ» |
idea that sparks
- dig deeper with deep research: one research -> research conception and topic -> deeper research and build the zoom-in and zoom-out effect
Context Engineering: The Key to AI Productivity
WIP -> new article to write
Trace
- 2025-04-20: initial version
- 2025-05-10: upgrade some more details and scenarios for AI to use and mapping to the tools and models
- 2025-05-23: add
AI Working Procedure
for better understanding of the AI working procedure and how to use the tools and models - 2025-09-05: upgrade to the recent tools / platforms / services