Latest / AI Fire Daily / #106 Robin: Running GPT-6 Astra & Claude Fable 5.1 Side-by-Side Free

Transcript

#106 Robin: Running GPT-6 Astra & Claude Fable 5.1 Side-by-Side Free · AI Fire Daily

• 0:00
  You know, you spend three weeks building the
• 0:02
  absolute perfect custom workflow. Oh, yeah. You
• 0:05
  upload all your specific brand guidelines. You
• 0:07
  refine your system prompts. Right. You load up
• 0:10
  your repositories, and it finally works beautifully.
• 0:12
  You feel incredibly productive. Then, maybe just
• 0:16
  a few days later, a totally new model drops.
• 0:19
  Yeah, exactly. Like Claude Fabel 5 .1. Right.
• 0:22
  Or maybe GBT6 Astra. It is objectively smarter.
• 0:25
  It is faster. It is fundamentally better. Yeah.
• 0:28
  But to actually use it, you got to rebuild your
• 0:30
  entire ecosystem. That is the hidden cost. Every
• 0:34
  new AI model feels like a brilliant magic trick.
• 0:37
  Until the excitement instantly fades. Beat. Welcome
• 0:41
  to the deep dive. Today we are exploring a radical
• 0:44
  shift in digital workflows. We are looking at
• 0:46
  a framework designed to prevent this exact type
• 0:49
  of burnout. Right. We are going to examine the
• 0:51
  trap of constantly rebuilding your workspace.
• 0:54
  Then we will introduce a unified local solution
• 0:57
  called NAMI. And finally, we will walk through
• 0:59
  a really fascinating real world test. Yeah, comparing
• 1:03
  four different models on a single task. Beat.
• 1:08
  So... The current reality of working with AI
• 1:10
  is incredibly messy. Well, absolutely. I mean,
• 1:14
  AI coding agents are practically everywhere right
• 1:16
  now. Yeah. These are AI programs that write,
• 1:20
  edit, and test code for you. They are powerful.
• 1:22
  Undeniably. But they create this deeply fractured
• 1:25
  ecosystem. You might find that ClogFable 5 .1
• 1:28
  is perfect for debugging. Right. But then GPT
• 1:31
  -6 Astra handles your creative formatting beautifully.
• 1:35
  Exactly. And then you still have Codex, Grok,
• 1:38
  or Kimi to consider. Because they each handle
• 1:39
  highly specific edge cases much better. Yeah.
• 1:42
  But honestly, juggling them all in a browser
• 1:44
  is just exhausting. I still wrestle with losing
• 1:46
  my project context myself. Every time I switch
• 1:49
  to a new model, you just feel... Like you're
• 1:50
  starting from a totally blank slate. Yeah, exactly.
• 1:53
  Well, you're... Definitely not alone in feeling
• 1:54
  that. Think about what happens when you open
• 1:56
  a new web interface. The AI knows absolutely
• 1:58
  nothing about you. Right. Or your current business.
• 2:01
  It demands that you explain your entire operational
• 2:03
  context all over again. You have to reupload
• 2:06
  your files, your assets, your historical data.
• 2:10
  You find yourself pasting your custom instructions
• 2:13
  over and over, moving your work between these
• 2:15
  isolated tools is just wildly inefficient. A
• 2:19
  modern business simply cannot afford to restart
• 2:21
  constantly like that. No, not at all. So the
• 2:24
  proposed solution in this guide seems relatively
• 2:27
  straightforward. You create one unified workspace,
• 2:30
  but you use many different models. Right. The
• 2:33
  actual underlying work remains exactly the same.
• 2:36
  This brings us directly to the concept of NAMI.
• 2:38
  Yes. From what I gather, NAMI is a free open
• 2:41
  source AI workspace. And the crucial detail is
• 2:45
  that it runs locally on your computer. Yeah.
• 2:47
  That local aspect is the entire foundational
• 2:50
  shift here. Instead of your work living inside
• 2:52
  a proprietary browser window, it lives locally.
• 2:55
  Right. It gives you one clean centralized place.
• 2:57
  to manage everything securely. You can actually
• 2:59
  run completely different AI agents right side
• 3:02
  by side. I was trying to visualize how this functions
• 3:04
  in practice. Think of it like having one highly
• 3:08
  organized master kitchen. That kitchen represents
• 3:11
  your main workspace with all your specific tools.
• 3:14
  And the different guest chefs stepping in are
• 3:17
  the various AI models. I love that analogy. They
• 3:20
  come into your kitchen to cook your specific
• 3:22
  meals, but they use the exact same ingredients
• 3:25
  and the same recipe book. Right. You are not
• 3:27
  building a brand new kitchen for every single
• 3:29
  chef. Exactly. That is a solid way to look at
• 3:32
  it. Yeah. Let me push it just a little further.
• 3:34
  OK. It is not just that they are using the same
• 3:36
  basic ingredients. The AI models are basically
• 3:40
  stepping in and using your exact muscle memory.
• 3:42
  Right. The ingredients and the specific techniques
• 3:45
  stay completely rooted in your system. The bigger
• 3:48
  philosophical idea here is making your business
• 3:52
  truly model agnostic. Because these models are
• 3:55
  just going to keep changing and evolving. Exactly.
• 3:57
  They keep leapfrogging each other. One specific
• 3:59
  model may absolutely win at coding tasks today.
• 4:02
  And another totally different model becomes vastly
• 4:05
  better by tomorrow afternoon. Does this mean
• 4:09
  the core philosophy is actually about separating
• 4:11
  your business logic from the AI engine itself?
• 4:14
  Yes, but it goes even deeper than just basic
• 4:17
  separation. It is really about taking ultimate
• 4:20
  ownership of your digital sovereignty. Right.
• 4:22
  The real functional system lives entirely in
• 4:25
  the parts that last. Your specific formatting
• 4:29
  instructions and brand guidelines are permanent.
• 4:32
  Yeah, your unique skills, local files, and creative
• 4:35
  assets. Completely permanent. Those workflows
• 4:37
  belong to you, not to some tech giant's ecosystem.
• 4:42
  Once those core parts are organized, any newly
• 4:45
  released model just plugs in. It immediately
• 4:47
  starts working from your deeply established base.
• 4:50
  You completely stop rebuilding your entire operation
• 4:52
  around every new model. Right. You just start
• 4:54
  running better and faster models through your
• 4:56
  existing system. So the AI is just a temporary
• 4:58
  engine. Your files are the permanent vehicle.
• 5:00
  Exactly. The engine gets swapped out when a faster
• 5:02
  one is built. But you never actually change the
• 5:05
  car you are driving. Yeah. B, I want to transition
• 5:08
  from this abstract philosophy to the actual mechanics.
• 5:11
  Sure. Setting up this NAMI workspace starts with
• 5:14
  one surprisingly simple move. You just point
• 5:17
  the software to your main AIOS folder. Right,
• 5:20
  and that folder becomes the absolute anchor of
• 5:22
  your entire system. You just create a master
• 5:25
  folder somewhere on your local hard drive. What
• 5:27
  kind of files are we actually putting in there?
• 5:29
  Basically everything you need to execute your
• 5:31
  daily work. It can include your rough content
• 5:34
  drafts and your visual thumbnails. It holds your
• 5:37
  video project files. crafted LinkedIn posts,
• 5:41
  short form scripts, client assets. And your specific
• 5:43
  brand colors. Yeah. But most importantly, it
• 5:47
  contains your system prompts and custom instructions.
• 5:50
  And once you open that specific folder in NAMI,
• 5:52
  everything loads automatically. Right. It pulls
• 5:54
  your entire directory right into a clean left
• 5:57
  sidebar. You can browse through your folders
• 6:00
  just like a normal computer workspace. And that
• 6:02
  visual familiarity is actually crucial for maintaining
• 6:05
  a fast workflow. You do not have to learn some
• 6:08
  bizarre new file management system. You can just
• 6:10
  open your content folder easily and check your
• 6:12
  thumbnail asset. Exactly. You can quickly look
• 6:15
  through complex YouTube project files. You can
• 6:18
  manually review any text document before ever
• 6:20
  asking an agent to read it. But for the AI to
• 6:23
  actually read those local files, you need MCPs.
• 6:26
  Yeah, that is a vital piece of the underlying
• 6:29
  architecture. Right. OCPs stand for Model Context
• 6:33
  Protocols. which sounds incredibly dense, but
• 6:36
  really, they are just rules that help AI safely
• 6:39
  connect to your local files. They act like a
• 6:42
  secure bridge. Exactly. A secure bridge between
• 6:44
  the AI's brain and your hard drive. If your skills
• 6:47
  and MCPs are already set up, everything just
• 6:50
  connects seamlessly. So because of that bridge,
• 6:52
  the agents pick up your brand tone automatically.
• 6:55
  Right. They learn your specific project rules
• 6:58
  without you ever having to ask. They access your
• 7:00
  hidden assets almost instantly. Exactly. And
• 7:03
  here's where the workflow actually becomes incredibly
• 7:05
  powerful. Inside NAMI, you just click New Session
• 7:08
  to start a task. And choose your desired model
• 7:11
  from a simple drop down menu. Right. You can
• 7:13
  start a coding task with Claude Code right away.
• 7:15
  Then open Codex right beside it in a separate
• 7:17
  panel. Exactly. You can easily add Grok, Kimi,
• 7:20
  or GPT -6 Astra to the mix. You basically load
• 7:23
  up any available model you want to experiment
• 7:24
  with. Whoa. Beat. Imagine running five completely
• 7:29
  different world -class AI models side -by -side
• 7:33
  on a single local folder. I know. It is wild
• 7:36
  to actually see it working. But wait, I have
• 7:38
  to challenge the mechanics of that setup. OK.
• 7:40
  Go for it. If you have one agent checking complicated
• 7:43
  code in the background and another agent is actively
• 7:46
  reviewing a thumbnail in that same folder, how
• 7:49
  do they not trip over each other's context and
• 7:51
  get confused? That is a really great question
• 7:54
  about how the memory actually functions. Yeah.
• 7:56
  You might think they would override each other's
• 7:58
  temporary data. Yeah. But NAMI actually isolates
• 8:02
  the working memory state of each individual session.
• 8:06
  Each session runs completely separately from
• 8:08
  the others on a technical level. Okay. But all
• 8:10
  of them maintain a secure read -only connection
• 8:13
  to the same workspace. So they can look at the
• 8:15
  data, but they cannot accidentally break it.
• 8:17
  Right. You can have one advanced agent working
• 8:19
  deeply on backend code. And another model simultaneously
• 8:22
  checking written content for brand tone. Exactly.
• 8:26
  And another could be testing a highly specific
• 8:29
  thumbnail task right nearby. they share that
• 8:32
  base OS folder and all its instructions automatically.
• 8:35
  But their internal scratch pads are completely
• 8:37
  walled off. Yeah, so they never interfere. Got
• 8:40
  it. Separate sessions, but they all drink from
• 8:42
  the exact same data well. Yeah, exactly. And
• 8:44
  because they drink from that well, the friction
• 8:46
  completely disappears. Right. Instead of starting
• 8:48
  from a blank chat window every single time, they
• 8:51
  start from your... fully realized operating system.
• 8:54
  It sounds really great in theory to have this
• 8:56
  one master folder, but theories often fall apart
• 8:59
  when you actually push them in reality. Oh, for
• 9:02
  sure. The guide outlines a tangible real -world
• 9:05
  case study to test this out. Yeah. They call
• 9:07
  it the thumbnail test, and it is actually quite
• 9:10
  revealing. This test shows exactly why having
• 9:13
  that shared local workspace matters so much.
• 9:15
  The setup is simple. Yeah. Inside the local AIOS
• 9:18
  folder, there is an existing thumbnail. And the
• 9:21
  goal for all different AI models is exactly the
• 9:24
  same. Right. They need to create a slightly modified
• 9:26
  version of that exact thumbnail. They have to
• 9:28
  replace a Claude logo image with a ChatTPT logo
• 9:32
  image. Yeah. And they must use the correctly
• 9:34
  specified brand assets located in the folder.
• 9:37
  So the exact same text prompt goes into four
• 9:40
  different active sessions. It goes to Kimi, Grok,
• 9:43
  Codex, and ClaudeCon simultaneously. Same exact
• 9:46
  task. Same local folder. Same background instruction.
• 9:49
  four totally different models. The main difference
• 9:53
  to pay attention to here is the starting context.
• 9:56
  In a normal browser -based AI chat window, every
• 10:00
  single model starts empty. You have to manually
• 10:02
  upload those logo files over and over again.
• 10:04
  Yes. You have to explain the nuanced brand guidelines
• 10:06
  repeatedly. You paste the complex formatting
• 10:09
  instructions again and just hope it understands.
• 10:11
  But inside this NAMI workspace, all of that rich
• 10:14
  context already lives there. Exactly. The various
• 10:17
  agents read the same local files effortlessly
• 10:19
  through the MCPs. They seamlessly follow the
• 10:22
  same underlying skills without needing a fresh
• 10:24
  setup. Let us look closely at the actual results
• 10:27
  from this specific test, because they are definitely
• 10:29
  not identical, which is fascinating to analyze.
• 10:32
  Oh, absolutely. Kimi actually completes the assigned
• 10:35
  task, but it makes a strange choice. Yeah, it
• 10:39
  inexplicably decides to keep the old color style
• 10:41
  from the previous thumbnail. Right. It clearly
• 10:43
  needs one more specific instruction to fully
• 10:46
  grasp the visual change. Yeah, that tells us
• 10:49
  something about how Kimi weighs historical pixel
• 10:51
  data. Exactly. It might prioritize the existing
• 10:55
  image structure over the new text prompt. Now,
• 10:58
  Grok performs a bit differently. Yeah, it actually
• 11:00
  handles the protocol better. It uses the MCP
• 11:04
  correctly to pull the right asset during the
• 11:06
  process. returning a very clean version featuring
• 11:09
  a black chat GPT logo. Right. But then Codex
• 11:12
  takes another visual path entirely with its output.
• 11:15
  It creates a perfectly usable version, but with
• 11:18
  a white logo instead. Which is interesting because
• 11:21
  maybe it's training data biases towards higher
• 11:23
  contrast designs. And finally, Claude Code returns
• 11:26
  something that looks very close to the Codex
• 11:29
  version. the visual results definitely vary across
• 11:32
  all four of the testing agents. Since the results
• 11:34
  vary, doesn't this prove that the model you choose
• 11:36
  still heavily dictates the final product? I think
• 11:39
  it dictates the specific flavor of the final
• 11:42
  product, yes. The internal reasoning engine of
• 11:45
  the model still ultimately controls the final
• 11:48
  quality. But the baseline competency is guaranteed
• 11:51
  by your local folder structure. Completely guaranteed.
• 11:54
  The massive, exhausting friction of setting up
• 11:57
  the test is entirely gone. Right. You can effortlessly
• 12:00
  test four different agents on the exact same
• 12:04
  complex task. You compare their nuanced visual
• 12:06
  outputs right side by side instantly. And then
• 12:09
  you simply just keep the one version that works
• 12:11
  best for you. Exactly. You basically run a real
• 12:13
  -time audition and just hire the best result.
• 12:15
  That's perfectly said. And this brings us to
• 12:17
  the ultimate conclusion of this entire guide.
• 12:20
  Which is? The absolute best AI setup must be
• 12:23
  able to keep improving over time. It should never
• 12:25
  force you to rebuild everything when a new model
• 12:28
  drops. Right. Nami also adds some nice quality
• 12:31
  of life features to make this easier. Oh, definitely.
• 12:33
  You can preview your local files easily right
• 12:35
  inside the main interface. You can pin essential
• 12:38
  visual assets to the screen while the agents
• 12:40
  work. Yeah. You can seamlessly switch between
• 12:43
  different visual themes in the settings. It just
• 12:45
  makes the entire digital workspace feel much
• 12:48
  easier to manage daily. But the bigger lesson
• 12:50
  here is profoundly simple. Yeah. You have to
• 12:53
  build around your own skills. and your own workflows
• 12:56
  first. Establish your custom instructions and
• 12:58
  your local assets as the permanent foundation.
• 13:02
  Build a system that makes sense for how your
• 13:04
  specific brain actually works. Exactly. incredibly
• 13:08
  strong model released in the future can just
• 13:10
  plug right in. If you want your modern business
• 13:12
  to easily keep up with GPT -6 Astra, if you want
• 13:16
  to seamlessly integrate the upcoming Claude Fable
• 13:18
  5 .1, if you want to casually use Codex, Grok,
• 13:23
  Kimi, and whatever comes next, you absolutely
• 13:25
  have to start making your daily workflow model
• 13:27
  agnostic right now. I highly recommend trying
• 13:30
  to test Nami with your own local AIOS. Just spend
• 13:33
  an afternoon organizing your messy files into
• 13:36
  one coherent local directory, and just see how
• 13:39
  profoundly different your daily creative work
• 13:41
  actually feels. It is genuinely remarkable when
• 13:44
  every new model runs from the exact same base.
• 13:47
  You stop feeling like you are constantly fighting
• 13:50
  the technology just to start working. Absolutely.
• 13:53
  There is something incredibly liberating about
• 13:55
  taking back that level of control. But I want
• 13:59
  to leave you with a final slightly provocative
• 14:02
  thought to ponder. OK. We established that your
• 14:05
  custom instructions and local assets are the
• 14:07
  permanent OS. And we established that these constantly
• 14:10
  evolving AI models are just temporary plugins.
• 14:14
  But what exactly happens when the AI models become
• 14:17
  truly autonomously advanced enough? What happens
• 14:20
  when the AI starts silently updating and rewriting
• 14:23
  our OS for us? Wow. Two sec silence. Thank you
• 14:27
  so much for joining us on this deep dive today.
• 14:29
  It was a pleasure. I really encourage you to
• 14:31
  look closely at your own digital workspace this
• 14:34
  week. Ask yourself a very critical, potentially
• 14:36
  uncomfortable question about your current setup.
• 14:39
  Is your daily workflow truly securely model agnostic?
• 14:43
  Or are you just temporarily renting digital space
• 14:45
  inside a massive AI ecosystem? It is a great
• 14:48
  question. Every single time a new model drops,
• 14:51
  it feels like a stunning magic trick. Do not
• 14:54
  let the exhausting setup cost completely break
• 14:56
  your creative workflow. Build your permanent
• 14:58
  foundation today and let the AI models work for
• 15:01
  you.