Take a couple hours to walk CC through your code and generate a CLAUDE.md. Note any architecture patterns you have already, or want to have, in your project.
This is probably the most important thing you can do to drive better results. As you work, try to ensure you're getting independently testable steps as you solve a problem. Take time planning, always have it reference your CLAUDE.md and existing code patterns. At the end of each step, I have CC determine whether or not to update the CLAUDE.md if there's any foundational updates.
The trick is to have a idea of what you're expecting out of these tools. If you can use the tool to break down the work into individual pieces you will find it is really fun and productive way to build software. You still have to think, but you are able to cover a lot more ground faster. I can't type out 4 files that are in my brain in 10 seconds.
I would much rather just contribute to the AI directly. I am kind of tired having to use so many different tools. I don’t even want to login or authenticate with dozens or hundreds of them.
Why won’t these companies completely transform how we work/vibe by having everything go through the chat/voice ui? It should just work, for the end user.
I’d love to be an agnostic business analyst for OpenAI or Anthropic. Used to be a business analyst and I loved understanding as much about an industry or domain and then working with stakeholders to come up with solutions.
Imagine if there were teams of business analysts internally at these companies to build out that super app.. supply chain (procurement, resolutions, dispatch), partner management, customer support, calling someone video or voice, all my documents, everything I keep basically. I want to work and communicate with other people over it. It should just work. All work is communication between people right?
I would like to know how much context is remaining. Claude code gives a % remaining when it is close to exhaustion which is nice, but I'd like to always see it.
Also, I wish it was possible for the models to leverage local machine to increase/augment its context.
Also, one observation is that Claude.ai (the web UI) gets REALLY slow as the conversation gets longer. I'm on a M1 Pro 32gb MacbookPro, and it lags as I type.
I really enjoy using LLMs and would love to contribute any feedback as I use them heavily every day :)
In my layman pov… A diesel engine can take the least aerodynamically shaped body and move it at 60 mph for 1k miles no problem. As an American, I guess it’s just natural to me that if it can move, then it should move with glory!!
The range only needs to cover the period between mandated brakes. Everything over that is wasted weight, as the weight of the cargo and the weight of the truck share a weight budget.
Yeah the range between mandated breaks is way over 100 miles. In the UK it is 45 minutes after 4.5 hours driving, so about 270 miles. But you need to account for the likelihood that there isn't a charger at the break stop. On the other hand the Volvo trucks in this article apparently have a similar range and they're selling so I guess it's worth the hassle.
That’s technically true, I suppose, but dishonest since you imply that those numbers are large enough to max out driving time. You wouldn’t be able to drive to your first mandatory 30 minute break with that range.
Many journeys will be from the distribution center to the supermarket or similar, or even multiple supermarkets, which would really bring down the average.
I should have been clearer about the specific problems I'm tackling. I’ll try addressing your points.
> On domain expertise
I've been in the trenches for over a decade since falling in love with Gantt charts, getting my CAPM certificate, and graduating college. First as a project manager on multi-year, $XX–XXX M projects, then as a self-taught SWE building PM software (we IPO'd a few years back). My domain knowledge is in oil & gas, construction, supply chain optimization, and manufacturing to a lesser degree. The scheduling tools I know best are MSP and Primavera P6.
> On the existing landscape
I agree that this is a well-established field of study. I'm using Google or-tools and evaluated a few other constraint solvers. However, I believe the focus has been more theoretical because actual software and tooling that people use day-to-day lacks any collaborative execution and management automation. In my opinion, this is where and why projects always break down.
* This leads me to the specific problem which I should have shared earlier! *
Every non-software rollout I was part of ran late and over budget. There's a stat that 90%+ of projects miss their targets. By contrast, agile software teams shipping sprint-sized features almost never slipped. The difference? Real-time collaboration and adaptability. (And yes, I'm aware the scopes are radically different!)
The moment we move from "just" the world of bits to the world of steel, gas pipelines, trains, and dealing with hundreds or thousands of people, equipment, 3rd parties... even the best plans crumble.
> Real world experience
I could go into minute detail on each one, but I'll spare you the reading and myself the PTSD by condensing a few pain points:
- No shared, real-time view. Data lived “somewhere,” but never with field teams or execs.
- Black-box auto-levelers (like Primavera P6) only delay tasks they don’t split or reallocate crews without the PM doing a ton of work. Small hiccups can easily compound into slips that last days or even months. Licenses are expensive and access is gated so only “the scheduling team” sees the live plan. I’d argue it’s not a solved problem otherwise we would not have needed to also bring on dedicated scheduling consultants.
- Process workshops and Excel jockeying can’t keep pace with real-world churn. Any good plan evaporates once work actually starts. We did these things on every project but I don’t think it really made a difference because while people are a variable, bad tooling is exponential.
> My approach
Rather than building another optimization engine, I’m creating a collaborative platform where scheduling is a core feature. Most tools focus on the initial optimization but ignore the continuous re-planning that real projects require. Not everyone _needs_ the schedule obviously, but the schedule _definitely_ drives the project. I want to enable realtime ADAPTATION as conditions/constraints change in the field.
I’m particularly interested in connecting with others who’ve hit similar walls, or who have insights on bringing real-time collaboration to complex, distributed projects so I really appreciate your response.
"9s" are brought up in reliability discussions. They refer to the number of 9's in an uptime metric like 99%, 99.9%, 99.9999%, ...
See also "march of the 9s"
I use pg and test suites run in parallel against their own db. It’s pretty cheap with CREATE DATABASE helpers plus using container means little to no headache.
Take a couple hours to walk CC through your code and generate a CLAUDE.md. Note any architecture patterns you have already, or want to have, in your project.
This is probably the most important thing you can do to drive better results. As you work, try to ensure you're getting independently testable steps as you solve a problem. Take time planning, always have it reference your CLAUDE.md and existing code patterns. At the end of each step, I have CC determine whether or not to update the CLAUDE.md if there's any foundational updates.
The trick is to have a idea of what you're expecting out of these tools. If you can use the tool to break down the work into individual pieces you will find it is really fun and productive way to build software. You still have to think, but you are able to cover a lot more ground faster. I can't type out 4 files that are in my brain in 10 seconds.