Hey @brookr 👋🏾

First, thank you for the very enlightening presentation!

AI tools can be used to write copy for support articles and other forms of content. They can be a tool to help start and even simplify that process. However, if you are a writer and have your own "voice" that's infused in the content that you create, AI tools cannot replicate that. There's also a level of context that can be specific to your industry, company, audience, etc, that AI tools can miss as well.

Question: What do you do that AI can't?

• AI coding assistant's struggle to produce code for uncommon tech stacks that may be unique to a company
• AI can't determine what needs to be created with out being told
• Empethize with the developer writing the code
• Completions lack domain knowledge, especially mathmatical knowledge
• Can't fully translate requirements from ley people who request code
• Locked into the context available
• Don't know the context of the viewer (may be overloading the user)
• Doesn't know when it is hallucinating
• How to reason about maintainability
• Can't implement features with all the requirements as once (limited to small snippets/single functions)
• Modify or update sections of code base that may be affected by other updates

What a developer can do that AI can't

Ideas discussed at GitHub Universe discussion lounge session:

• Consensus - gather people together, understand what we want built
• Prioritizing work that aligns with your vision. Big picture planning.
• Business logic - unique scenarios and requires customization
• Domain-driven development & design
• Bringing diverse perspectives - when trained on their own output, LLMs return garbage. Multiple LLMs aren't discussing and innovating together based on diverse perspectives. All trained on existing code, not innovating anything new or pushing boundaries.
• Copilot/AI has been observed writing bad code.
  • Good prompting is important and can help
• A human must still review code. Currently looking at it as more of a reference that must be reviewed and changing it to get it up to meet standards before accepting it.
• How do you reconcile AI with test-driven development? Writing tests is not difficult, knowing what to test is. Having AI write tests for you can become a self-validating practice.
• AI can't determine or understand "why am I building this?"

"What do you do that AI can't?"

Notes

• Understand the system and codebase as a whole
• Understand customer requirements
  • This could get better
• Understanding the problem
• Talking to people - it is a superpower
• Ask the question "are you sure you want that?"
• Creativity
• Decide which solutions are correct or incorrect
• Ensure the correctness of solution
• Think at a higher level - why are we doing this in the first place? Do we need to create this database?

AI takes longer to learn

Understand requirements
(read between the lines --- low context cultures)
Natural language requirements are sparse and require interpretation

Need for software engineers keeps increasing.

• can complain about stuff
• i know a lot of bio nuance and can use copilot to write code but i know context eg, certain types of genes must be excluded. chatgpt is good but some domains its not well versed at
• my work involves talking to people and mapping business
• the range of projects is crazy diverse, hard to create one particular project because wont have applicability to next project
• ideation -- where do we want to be in 5 years -- its about predicting, its about shaping
• product that copyrights IP, use standard AI classifiers to see if they dont recognize then it's an original problem
• good at known knowns, bad at unknown unknowns
• can ai feel? there's that too... how does that apply in the workplace? empathy
• situation that's too taxing, doesn't know individuals' work capacities
• opposite of user-centric design
• ai can't own a project
• if not on-prem, can't trust with org knowledge -- can't trust people either
• i can find problems in things
• pushing for organizational things

Q: What do you do that AI can't?

• Deal with incidents, incident response
• Have zoomed out, high level discussions, white boarding process diagrams and translating that into code
• Juggle priorities with stakeholders; stakeholder management
• Onboard new developers, knowledge transfer
• Come up with outside-the-box, non-standard solutions that are heavily nuanced
• Be continuously aware of context and the bigger picture
• Gather user requirements
• Have people skills

What do you do that AI cant

We listed things that AI could not do that we could

• AI cant bypass limitations in your organization (tools, understanding infra)
• AI can't provide context without the for your organization and
• AI can't map your business needs directly to code; you need additional context
• AI is generic it doesn't quite understand everyones context
• AI is not great at modern stuff and recency; it can also hallucinate
• AI cant distinguish facts from fiction (LLM specifically)
• AI can't dig into things at runtime and pull out the type of problems that lead to optimization's
• I can translate user needs into technical solutions; AI often struggles to do this or takes it too literally
• You have to validate and build things from a human perspective; AIs cannot validate themselves
• AI can't respond to my emails

Here's the summary:

Decision Making and Judgement
A. Innovation and Vision

1. AI cannot foresee market trends and innovation.
2. AI lacks the ability to make initial product decisions.
3. AI cannot predict or shape long-term business strategies.

B. Understanding and Empathy

1. AI struggles with comprehending psychological differences.
2. AI cannot empathize or understand nuanced human emotions.
3. AI is unable to manage personal workloads considering individual capacities.

C. Business Acumen and Direction

1. AI lacks domain expertise and the ability to understand business directions.
2. AI cannot autonomously maintain a consistent set of principles.
3. AI is unable to drive domain-driven development and design based on unique scenarios.

II. Technical Proficiency and Creativity
A. Coding and Development

1. AI relies on prompts and cannot innovate new code independently.
2. AI requires human review to ensure code quality and standards.
3. AI cannot fully understand or test complex software systems.

B. Problem-Solving and Creativity

1. AI lacks the creativity to develop outside-the-box solutions.
2. AI cannot understand high-level systemic problems or the 'why' behind projects.
3. AI struggles with ideation and predicting future needs or trends.

C. Adaptation and Learning

1. AI cannot learn from or understand complex and diverse project ranges.
2. AI is not equipped to onboard new developers or transfer knowledge effectively.
3. AI has limitations in adapting to organizational tools and infrastructure.

III. Communication and Collaboration
A. Interpersonal Skills

1. AI cannot handle interpersonal situations like dealing with angry people.
2. AI lacks the ability to gather consensus or bring diverse perspectives together.
3. AI is not capable of managing stakeholder relationships or user-centric design.

B. Contextual Understanding

1. AI struggles to read between the lines or interpret sparse natural language requirements.
2. AI cannot effectively translate user needs into technical solutions without additional context.
3. AI is unable to juggle priorities based on a continuous awareness of the bigger picture.

C. Organizational Integration

1. AI cannot own a project or drive organizational changes.
2. AI is not capable of dealing with incidents or emergency responses.
3. AI cannot be trusted with organizational knowledge to the same extent as humans.

Let's see that as a mind map: