Daniel Barcellos

Building in public
Payments, systems and distribution
Engineer in Silicon Valley

[Daniel Barcellos @danbcellos]

These years in the industry I learned the problem is not software. Everyone can create build anything

But can we

observe anything

keep anything up 24/7

respond, react and predict anything reliably?

Now looking at how oil and industry plants think their reliability and quality control, It’s more clear that we’re not a framework or bubble away to solve the problem of keeping having customers

In fact we have decades of math that proves software building is just the tip of the iceberg

[Daniel Barcellos @danbcellos]

Google sre philosophy sais two things :

- Every alert should have immediate actions
- We should alert on symptoms not causes

Do you agree?

[Daniel Barcellos @danbcellos]

@ku_ds17868 I mean, everything , but videos is the hardest so far

[Daniel Barcellos @danbcellos]

Do you really get along with AI generated content?

[Daniel Barcellos @danbcellos]

I never thought uncanny writing was a thing before I tried creating video scripts with ChatGpt

[Daniel Barcellos @danbcellos]

@MihaiButnaru Please!

[Daniel Barcellos @danbcellos]

@Thomas_Tao_1 But does it hit sometimes with writing generally with those minor improvements over human touch

[Daniel Barcellos @danbcellos]

@Thomas_Tao_1 Yess

[Daniel Barcellos @danbcellos]

@ku_ds17868 In a sense that you use that for the script improvement or also video image as well?

[Daniel Barcellos @danbcellos]

@MihaiButnaru Teach me haha

[Daniel Barcellos @danbcellos]

@ku_ds17868 How do you do it?

[Daniel Barcellos @danbcellos]

Very direct

[Michael Trinh @_mctrinh]

The skill ceiling in software engineering is moving upward because the floor keeps getting automated.

[Daniel Barcellos @danbcellos]

@bymarvellous That’s it!

[Daniel Barcellos @danbcellos]

Replication and team work compounds

[Daniel Barcellos @danbcellos]

Netflix built Chaos Monkey to randomly break production systems and test resiliency.

What would a Chaos Monkey for AI look like?

Injecting hallucinations into live pipelines to see if your agents, guards, and humans can actually catch them?

[Daniel Barcellos @danbcellos]

Use AI to find the deterministic solution you need or you’re one hallucination away from loosing trust

[Martin Fowler @martinfowler]

Fragments: thoughts from a software development retreat, would I be a writer without text editors, use LLMs as functions, don't use LLM skills, is the future of everything lies?

https://t.co/M1V3vOytjR

[Daniel Barcellos @danbcellos]

I felt that first 2023 with the introduction of Jaseci and Jarvis

[Uncle Bob Martin @unclebobmartin]

I hold a different opinion. I think he’s right about things like domain driven design, bounded context, ubiquitous vocabulary, and the expression of the conceptual domain. I think he’s wrong that computer languages (i.e. languages of procedure) will be the vehicle of that expression.

In my mind such languages will become the private domain of the AI’s. We, programmers, will use a different formalism. Something like gherkin, or some other formalized statements of specification. Still formal, still precise, but nothing at all like the sequence, selection, and iteration of computer code. The language we will use will not be procedural, or object oriented, or functional, or anything related to the individual steps executed by a computer. Rather that language will be a declarative specification language. A set of goals that the AI must achieve.

[Daniel Barcellos @danbcellos]

@unclebobmartin @plainionist How far from that are we?

[Daniel Barcellos @danbcellos]

Here’s what we learned hitting 700k+ views with 4 videos after testing multiple strategies:

• Talking head content → dead
• Educational/explainer content → low conversion
• Viral content alone → attention, no money
• Viral content + indirect product references → 400 hot leads + 10 recurring sales

The product doesn’t need to be the content.

It just needs to exist inside the attention.

[Daniel Barcellos @danbcellos]

True

[will (in sf rn) @ItsWillHenry]

the 5 most important things nobody tells you while building a company:

1. the co-founder relationship will determine everything before the product determines anything.

2. the first ten customers will tell you exactly what to build next if you actually listen to them

3. your burn rate is a clock and every bad hire makes it tick faster. slow down on people. speed up on everything else.

4. the pivot you're afraid to make is usually the one that saves the company. everyone around you already knows it. you're the last one.

5. nobody is coming to save you. not the investors, not the advisors, not the next feature. just you, the problem, and how long you're willing to stay in the room with it.