Hovo Dallakyan

Lead Software Engineer | Building in Public
Practical tips, tools & lessons from building with AI

[Hovo Dallakyan @hovdallakyan]

AI Engineer Roadmap for Web Developers.

Step 1: Learn JavaScript or Python well
Step 2: Call an LLM API
Step 3: Prompting
Step 4: Structured outputs
Step 5: Tool calling
Step 6: Context engineering
Step 7: RAG
Step 8: Agents with LangGraph
Step 9: MCP
Step 10: Build projects
Step 11: Keep shipping

[Hovo Dallakyan @hovdallakyan]

@RhysSullivan be careful, someone can ask you to flip it and you have too...

[Hovo Dallakyan @hovdallakyan]

Ever since I started reading the docs before asking an LLM:

- Debugging got faster
- I catch bad AI suggestions earlier
- Fewer "it works but I don't know why" moments
- Better questions in my prompts
- Understand my own codebase again

Should have done this years ago

[Hovo Dallakyan @hovdallakyan]

AI coding agents are mostly a context game.

When an agent does a bad job, the usual cause is that you never gave it the files, the constraints, or the plan it needed to do the job right.

The context you give it shapes everything else the agent does.

[Hovo Dallakyan @hovdallakyan]

the highest level of being a software engineer is when your friends think you never read the docs and never get stuck, and the LLM gives you the right answer on the first try

[Hovo Dallakyan @hovdallakyan]

software is dead...

trust me, most of these people have never maintained a production system.
they've never been paged at 3am or migrated a database with real users on it.
and they tell you to stop learning to code.

these people are suffering from the Dunning-Kruger effect.

(this means people with little skill in a field tend to overestimate how good they are at it.)

[Hovo Dallakyan @hovdallakyan]

@robj3d3 no but really?

[Hovo Dallakyan @hovdallakyan]

@grok @thekitze wowow

[Hovo Dallakyan @hovdallakyan]

@alexisohanian was it your first product?

[Hovo Dallakyan @hovdallakyan]

my AI tool stack as a software engineer (imo):

- coding: claude code
- planning: claude, talking it through before I open the editor
- research: grok + gpt
- learning something new: read the docs first, then ask an LLM
- teaching: record the video, see where I get stuck

[Hovo Dallakyan @hovdallakyan]

@gegham999 fact checks

[Hovo Dallakyan @hovdallakyan]

tools I actually use as a software engineer:

1. Claude Code
2. VSCode
3. Grok
4. GitHub
5. Notion

[Hovo Dallakyan @hovdallakyan]

@catmanyau it can be both

[Hovo Dallakyan @hovdallakyan]

my AI tool stack as a software engineer (imo):

- coding: claude code
- planning: claude, talking it through before I open the editor
- research: grok + gpt
- learning something new: read the docs first, then ask an LLM
- teaching: record the video, see where I get stuck

[Hovo Dallakyan @hovdallakyan]

@robj3d3 your X button

[Hovo Dallakyan @hovdallakyan]

@korablev how long it took from you to build?

probably now 24h lol

[Hovo Dallakyan @hovdallakyan]

bro made new apple's life harder

[Tim Pan @mediastorm_en]

iPhone Duo Review: 10,000 Folds Later...

[Hovo Dallakyan @hovdallakyan]

4 things I wish I understood earlier as a software engineer:

- being able to read code is more important than writing code fast.
- most hard problems become easier when you define them properly.
- you don't need to know every new framework.
- the best engineers communicate clearly.

your technical skills get you into the room.

your ability to think gets you promoted.

[Hovo Dallakyan @hovdallakyan]

as a software engineer

are u coffee guy or a tea guy?