The future of many jobs may seem uncertain, even in the Coeur d'Alene and Spokane area. 

Upon encountering ChatGPT, it's natural to feel apprehensive, but let's not forget the rooms full of human "computers" in the 1950s. Rather than succumbing to fear, let's predict where this technology is heading and develop skills and tools to harness the power of AI.



<div style="max-width: 420px; margin: 0 auto;">
     <img src="/assets/images/articles/midjourney_1.jpg" class="img-bordered">
     <br>
     <small><i>Some fun attempts to generate an avatar, using [Midjourney](https://midjourney.com/)</i></small>
</div>

The release of ChatGPT revolutionizes AI's role in business, offering our boutique 
software studio the opportunity to provide clients with innovative solutions. By 
harnessing ChatGPT, we can develop intelligent chatbots, enhance content generation, 
and create tailored AI agents to streamline clients' operations and elevate their 
digital presence.

# How we're using AI

### Chrome Upwork extension
Respond to job offers quickly by having an AI agent knowledgeable about your business generate proposals for you. Can answer questions like "what frameworks do you know?" or "talk about a recent project like this." Uses [OpenAI](https://openai.com/)

### Game development
We're leveraging AI to create artwork and large playable maps for our game. We doodle what the map should loosely look like and let AI fill in the rest.

### Website redesign
Building a fresh new design around [Midjourney](https://midjourney.com/).. the one you're browsing right now!

### Prompt Wars
A fun website using [InvokeAI](https://github.com/invoke-ai/InvokeAI) where users generate images that compete with others based on challenge topics.


## AI Agents

### FishingTravelAgent-GPT
   * For fun I made a quick fishing trip travel agent that helps you find the most interesting fishing spots on your journey

### JuniorEngineer-GPT
   * We're working on training an agent that can do basic coding tasks for us, like a junior engineer that's maybe Sophomore level in college
   * We can provide example tasks, ask for changes, and the agent will do its best to complete the task
   * This can be automated so the agent could look at new tasks, figure out how to do them, and push the code up


# How we can help you

Let's say you have a business that needs to do a lot of repetitive tasks, like
replying to customer support tickets, or maybe you need to generate a lot of
content for your website. We can help you create an agent that can do these
tasks for you.

We can also help you create an agent that can be your AI assistant. For
example, maybe you are annoyed by scheduling your employees every 2 weeks.. "Sally
doesn't get along with Bob, so we need to make sure they don't work together". We
can teach an agent about this, provide scheduling needs and requisite information, and
the agent will behave like a human scheduler!

Another application is in advertising. Compose a long list of straightforward ads, and let ChatGPT transform them into engaging, high-impact copy, optimizing your marketing efforts with ease.

Fun aside: in the past we helped architect the [Codalab](https://competitions.codalab.org/)
platform, which is a platform for hosting AI competitions. Our special piece
for that project was the ability to use "docker in docker" for student submissions.
This allowed AI research in C#, Julia, R, Python, etc.!




# Computers can see!?

Facebook (Meta) created the Segment Anything Model (SAM) using a dataset of 1 billion 
masks on 11 million images. This is an insane tool that can be leveraged quickly
to facilitate a strong AI agent.

<div style="text-align: center;">
     <img src="/assets/images/articles/SAM-model-Metas-new-Segment-Anything-Model-explained-4.webp" class="img-bordered">
     <small><i>Example SAM mask</i></small>
</div>

Maybe you have an assembly line where items are packaged, shipment labels are applied,
and then the packages are sent out. What if you could add cameras along the way
that help catch any potential problems?

Say you sometimes mix up some shipments, like shipping 1 BOX of items instead of 1 UNIT.
An AI agent could be trained to see the difference between a box and a unit.

Maybe sometimes a wheel gets stuck in the assembly line and the boxes start piling up,
an AI agent could see this problem and notify the appropriate people.

These ideas are not revolutionary on their own! The novel piece is how simple and quickly this can be done. 
Instead of needing a research team that can make all of the complex math work, we just
teach the AI agents like we teach a new hire. It's flexible and generic.


# Autonomous agents

A new thing coming out are Artificial General Intelligence (AGI) agents, like:
 - [BabyAGI](https://github.com/yoheinakajima/babyagi) - A baby AI that has limited memory and functionality
 - [ToddlerAGI](https://twitter.com/gogoliansnake/status/1643225698801164288/) - Modified baby agi, slightly more advanced
 - [TeenageAGI](https://github.com/seanpixel/Teenage-AGI) - Has a deep memory, more advanced than toddler
 - [AutoGPT](https://github.com/Torantulino/Auto-GPT) - attempt to make GPT-4 fully autonomous

# Let's bring your business into the AI future!

We're using AI right now with multiple clients. One client is developing a new
way to manage his pecan farm, leveraging [Github Copilot](https://github.com/features/copilot) 
along the way to speed up development.

ChatGPT and OpenAI: Unlocking the Potential of AI for Your Business

I certainly love most of my work, but my favorite part is pair programming, 
which is when a couple developers share a screen and work together solving 
problems, which facilitates spreading knowledge in a big way. 

It is extremely important for scaling Ckc that we teach new folks how to 
think -- a bit strange, but I'll explain!




<div style="text-align: center;">
     <img src="/assets/images/articles/decision-tree.png" class="img-bordered">
     <small><i>Why go up the tree until you're sure about the bottom!?</i></small>
</div>


# An example

1. You are tasked with setting up a new staging server in the Paris Headquarters
1. You set up the server and everything works, but, wait.. for some reason, connections are being denied to [Minio](https://min.io/) on port `9000`?!
1. You're a junior, so you're not sure how this is even happening. Everything
    is working on your machine? Everything else seems fine, just this one service is borked
1. The week of debugging begins:
   * The connection error, maybe it has to do with SSL at some level?
   * Maybe there's something wrong with my Minio configuration?
   * Is Minio booting up OK?
   * Is the port open in Security Groups? Yes..
   * Why can I talk to minio from the local machine, but the outside Internet can't talk to it?
   * I've tried 1,000 times and everything seems fine!
   * This goes on for days
   * Finally, I try a random new port and.. wallah! It works!


It turns out the organization hosting our server disallows port 9000! I had no
idea. I never even thought to ask.

Basically changing one number, a port, from 9000 to _anything else_ fixed the problem. Why
didn't I just focus on the initial connection error? Why did I go down so many useless
paths? I knew what the problem was at the outset, but refused to focus on it. 

I never expected a port to be closed by an outside admin, this had never happened 
before. I could see the list of open ports in my configuration, but I refused to acknowledge it clearly
wasn't open.

I couldn't connect: that's it! Focus on that! It _has_ to be a network problem, nothing
to do with Minio configuration.


# Let your mind go, but with limits

Confronted with so many unknowns you may feel hopeless. You have no idea what 
to try next. You, maybe literally, start to panic. Do you even belong here? Do you
even understand this stuff? You think it should be working, but it's not, and 
reality is crumbling around you. Trouble focusing, wanting to pull your hair 
out... all of that fun stuff.

Let this wash over you, maybe take a second to enjoy this feeling of not knowing
how to do something -- eventually this challenge will devolve into another way to
do a CRUD form stored away in your brain.

You'll forget what a rush it was to finally figure out you needed `levioSah()` not
`levioSo()`. This _will_ be boring someday, I promise, and you'll miss these days.


# Start at the base of the decision tree

After the initial panic has washed over you, re-examine the problem in the simplest
way you can. In the above example it was clear from the very beginning it was a 
connection issue.

Why did I go any further up the decision tree? I had a bad connection, that's it, 
maybe try to change ports or something?

It took a week to diagnose this port problem, which should have been diagnosed
in 5 minutes given a stoic "senior engineer" behind the wheel.


# Furthest branch away

Very rarely you will encounter a problem where all initial assumptions/clues are actually
futile and there's something fundamental wrong with your computer architecture
or the library you're using. Usually you did something silly like a typo, forgot to
pull, forgot to rebuild docker, etc.

Be very cautious going so far up the tree, re-check your initial assumptions. Re-read the 
error. Restart your computer.

Very likely you just need to open a port or something simple!


# It's not always easy

And, actually, sometimes you do have to sit there and take it for 2 weeks. Fully
absorbing an idea and figuring out a problem may legitimately be very difficult
and all-consuming.

Even though this port 9000 problem was silly and took way too much time to understand,
I learned a boatload. This always happens. Every time I am super stuck and find
the solution: I come out stronger, more apt to take on bigger tasks.

Over time, you get better and better at knowing when to climb up the tree and try 
to explore deeply.


# As a process

Here's a repeatable way to try to avoid going too far up the decision tree:

1. _Calmly_ read all errors, code, commands, etc. in front of you.
1. If something is obviously broken, focus on that. Do not assume a third party
system/module is at fault. It's probably you, and it's probably a very small simple problem!
1. Build a small example replicating exactly your situation with as few moving parts as possible _(i.e., from our example, open any kind of server process and serve it on port `9000`)_
1. Rubber duck your situation to a peer.
1. Reach out to someone smarter than you. If you're the top dog: God help you.
1. Now start posting on forums, discord, IRC, etc. and try to find any more info




# Past related articles
   - [How to get a web dev gig?](https://ckcollab.com/2019/03/30/how-do-i-get-a-web-dev-gig.html)

The Decision Tree

We've been constantly improving [Skeletor](https://github.com/ckc-org/skeletor/), our base template for new projects, over the years.

Most recently we added big chunks of functionality for mobile apps via React Native and
a standalone installation script that doesn't require django-admin on your command line.



<div style="text-align: center;">
    <img src="https://github.com/ckc-org/skeletor/raw/master/docs/skeletor_creation.gif" style="margin: 0 auto;">
</div>


### Fast install, standalone script

All you need is this command, typing in your project name, and selecting an option or two .. that's it!

```bash
$ bash <(curl -fsSL https://skeletor.ckcollab.com)
```

Now you have a Django + Vue + React Native project you can deploy to Heroku and get on the
app stores very quickly. You could add realtime chat in a few steps with [django-chit-chat](https://github.com/ckc-org/django-chit-chat).


### Mobile!

One of the biggest changes for Skeletor, and for Ckc in 2022, was adding mobile app support. We've steered
away from that ecosystem until recently, where we've found [React Native](https://reactnative.dev/) to be a pretty great tool.. 
mostly thanks to the folks at [Expo](https://expo.dev/) and their nicely packaged modules.

We played with [Native Base](https://nativebase.io/) this year and were actually burned pretty
badly by it. We started out with this framework, and it seemed fine, but pretty soon
we were experiencing horrible latency loading some components. Then we hit a final straw
where things were just too slow and had to rip it out.

Funnily we used ChatGPT to help replace a lot of these components! We would present 
the chatbot with our component code and ask it to be replaced with default React components. 
It was shockingly close most of the time, and did a perfect job often!


### Many small improvements

 * Built in login/logout/password reset request utilities you need on almost every project
 * Better handling of Heroku by default (Redis SSL anyone?!)
 * Various changes in `backend/settings/test.py` to make tests faster (cheap password hashing, etc.)
 * Better handling for emails/email templates
 * Django Migration checker before git pushes
 * Prevent use of `v-html` dangerously
 * Better Vue linting in general
 * Better S3 and Cloudcube support
 * Remove many django auth third party dependencies

### Does this shit even work!?

Has been used successfully to completion on over a dozen projects!

It is not unusual for us to get a new project in the afternoon and have a proof of concept or 
simple solution going by that night or the next day. We can move extremely quickly with
our toolset.


<div style="text-align: center;">
    <img src="/assets/blog/skeletor-upgrade/react-state.png" style="max-width: 420px; margin: 0 auto;">
</div>

### Upcoming changes

 * Add opinionated react support, once we have some good opinions!
 * More modules in the mobile base app, like password reset
 * And more broadly, semi-out of scope for this article, is more Ckc maintained packages, we currently have:
   * [django-ckc](https://github.com/ckc-org/django-ckc) - nice helpers for django in general
   * [django-chit-chat](https://github.com/ckc-org/django-chit-chat) - chat via websockets (channels!)
   * [dotfiles](https://github.com/ckc-org/dotfiles) - boot up a dev machine with neat presets

Skeletor gets some upgrades!

Heroku.. almost a bad word in many circles. It's expensive, polished, and
restrictive. To me: it's like the Apple of hosting providers. I can rely on it to
_"just work"_ the vast, vast majority of the time.



<div style="text-align: center;">
    <img src="/assets/images/articles/heroku-logotype-horizontal-white.png" class="img-bordered">
</div>

## Why Heroku?

**Managed hosting.**

As a developer first, I want to get right to the meat of the business logic and start
solving problems. Getting around configuring an EC2 instance for the
umpteenth time is an incredible productivity boost.

It's like having your own dev ops team. Right out of the box we can leverage many
managed services like automated deployments, backups, security updates,
quick recovery from outages, etc.

For example, instead of spinning up an instance of Postgres, making our own backup
solution, verifying it works, writing tools to make it easier to make backups before
deploys, etc… we can use Heroku Postgres which comes with WAL (Write Ahead Logging) aka
realtime live backups with a 4hour rollback window.

### Pros:

- Extremely friendly to developers, can onboard or transfer to new people very easily
- Familiar to many developers
- Mostly painless deployments
- Integrates with many services easy, ie CircleCI
- Load balancing for free

### Cons:

- Hard to customize if we have unique needs
- Relying on third parties that can go down (but, is based in the same facilities as
  Netflix)

<div style="text-align: center;">
    <img src="/assets/images/articles/aws-logo.png" class="img-bordered" style="background-color: rgba(255, 255, 255, 0.8);">
</div>

# Why AWS?

**Control.**

If you need some complicated configuration of Postgis, nginx caching, non-emphemeral
local storage, yadda yadda.. it's probably easier to go straight to EC2 instances
than it is plumbing some gross configuration on Heroku together.

Also, if you are running a VVIP top secret app then you will want to sandbox your code
away from any prying eyes. But, in my opinion, this is rarely needed and often times
for small-to-medium sized enterprises: security theater. Certainly it is another
point of failure to rely on a separate service storing your code, but I'd bet
on a managed host with hundreds of millions behind it.

### Pros:

- Complete control -- if we want to do some funky settings on the Postgres database and
  use it more like a graph database, add map projections, etc. then we can do that
- No third party outages to worry about
- No third party, beyond AWS itself, has access to your code

### Cons:

- More prone to human error, compared to Heroku with dozens of Dev Ops working full
  time to keep the pipes unclogged
- Requires a more experienced Dev Ops person to maintain
- Each setup is different, everyone has their own opinions and that means new devs
  take a bit longer to get spun up

# Conclusion

At CKC we love simplicity and speed. Heroku brings us the capability to develop
complex applications as quickly as possible, and to run those applications
at scale.

Have an app you're having trouble deploying? Let us [setup something for you!](mailto:hello@ckcollab.com)

ChatGPT and OpenAI: Unlocking the Potential of AI for Your Business

The Decision Tree

The Decision Tree

Skeletor gets some upgrades!

Skeletor gets some upgrades!

Heroku vs AWS

Heroku vs AWS