Terminal Bench 2.1 is saturated. Many token saving techniques would save money and score basically the same running Fable 5 against Terminal Bench 2.1. (They claim a better score but don’t say how much better. I’d bet my favorite hat that it’s not statistically significant.)

This is at least the fourth time I’ve seen a project hit front page with a “save money with same score on saturated benchmark” claim.

It's wild to me to claim that it's tricky to customize one of these harnesses and for that to be the entire justification for an entirely different harness.

It's really not that hard. If you want to reduce costs then all you need to do is practice delegation: instead of using the strong model, all the time to do everything, instead, you have the stronger model delegate well-defined tasks to a weaker model. Patterns like these are really easy to wire up.

Should give you some competition.

However, for this kind of customisation, Pi is actually quite great. One of the most things I love about Pi is ability to ask it to create an extension and it does it quite well as it’s part of their docs. Also ability to customise the system prompt to avoid the clutter that Claude Code add (around 20k system prompt that mostly had nothing to do with the code).

The demo was showing something I have created for my Pi setup, which is asking me in each new session which skills and MCP I would to enable for the session. This works quite well if you have multiple projects where you don’t need all skills but just a small subset

I wish this wasn't the case. I love software, but it's becoming increasingly easy to just build my own.

I would seriously rather see a plain unstyled html page with a marquee tag saying "A HUMAN WROTE THE NEXT THREE PARAGRAPHS"

I switch between Pi, Autolith, and my own harness. I come to this website. I don't know what the hell a strand agent is. I don't know who is putting this out. I don't understand whether I'm meant to use a GUI or write code in order to configure it. All the screenshots show a GUI. "use one line of python or javascript to configure." I don't know if this is a harness, an orchestrator, an agent framework, or all of the above. I think it's hard. People get so caught up in making their own thing - they lose track of the state of the art. I have no idea what the value prop for Strands Harness is. But maybe I'm not the target audience.

Also, I don't really care about 30% token efficiency gains. Maybe I should. But this is not my experience. Don't get me wrong, I'm conservative with tokens. But I also have a LOT of convenience / essentials which bloat up context. And I'd rather bloat context and have it work well than not have it work well. And when GLM 5.3 Flash came out - all of a sudden I had more tokens than I knew what to do with. That's when I started using OpenClaw type agents for the first time.

I don't have the tokens/cash to throw at something like Gas Town or OpenRig. And I'm still not at all convinced that those software factories are actually the future of everything. Maybe so. But I just don't see why I would invest the time to learn a piece of software without really understanding the value prop. It's crazy to say, but it's just not enough anymore for software to be free. Because the cost of me implementing my own software is rapidly approaching free. And I like mine better.

"Oh, well I just need something that works out of the box." Wait, why? What does it need to work for? For making software? lol.

Setting up projects and working with the AI on local files has been great, but only for my personal account. I've been trying to get it set up for work that provides OpenAI models through a 3rd party tool, company hosted models, as local-machine models and the UX is just straight up awful.

there's no GUI for profiles or custom endpoints, the config.toml sucks and the overall experience is primitive.

At this stage I really want a Codex-like harness but I need more fluid control over the models, I want features like pinning a project to a provider, as well as pulling in all the models from that provider, also having all providers available.

So if I need to pop over to one project to consult about product A, then pop to another project to do some code analysis on product B I can do so fluidly and have my tokens billed to the right place for each concern.

or a project that can span all of the resources. like having the OpenAi models orchestrate sub-agents on the local or hosted models.

The big threat to AWS is that coding agents dont need all of their complicated infrastructure, which was built for humans. Agents can use low level primitives, i.e. just a raw server

But the moment you build your own agent, you’re on your own. It’s tricky wiring up the right primitives just well enough to match that “it just worked” feeling."

Im sorry, but who is saying this? If you just throw this statement into agent of your choice- and ask what native integrations exist to cover this use: OAI and Anthropic both have a handful of options here. Claude Agent SDK, Claude managed agents, Codex exec, Codex sdk, Codex app server, openai agents sdk, openai agents api.

Ive heard from a 25 yoe consultant in a meetup group in person that aws agentcore was THE best way to handle enterprise agentic workflows with all of the proper knobs for governance etc since it comes with the iam integrations and arns etc.