Sabitlenmiş Tweet
Stuart Chaney
1.7K posts
Stuart Chaney
@stuartchaney
Founder/CEO @RivoCommerce. Powering retention programs for DTC brands like @HexClad, @kitsch, & @ridgewallet. Bootstrapping to $25M+ ARR in ecom SaaS
Rivo + HexClad case study 👉 Katılım Ocak 2023
498 Takip Edilen2.2K Takipçiler
Stuart Chaney retweetledi
i'm going to give some free saas advice here
with analytics the data pipeline/sync/getting the numbers to be correct is not trivial. rest is just fancy js charts on top of a db
dealing with all the edge cases with the usual data sources takes 6 months to a year. getting it to 98% is relatively easy, but fixing the rest of it takes another year and then you're already hitting scaling issues with your first enterprise customer with million+ orders asking their metrics to not have one day lag while hundreds of🥦 hair dropshipper demons keep pinging you for that one order or metric not matching with shopify/fb/whatever other integrations you offer, as the whole brainrot generation has the same attention span as my 4 year old - they are simply incapable of waiting for a minute for the latest order to show up in the dashboard
most established providers are for sure overcharging but you will make a lot more $$$ by just focusing on agency work rather than shipping mediocre saas when no one is even acquiring analytics apps anymore, with klaviyo trading at sub 3x ARR.
the market is already saturated and dozens of analytics apps are being launched every month
good luck with the path you've chosen, i hope this rant saves someone else at least from the honeypot idea of building your own analytics app only because TW overcharges
Collin Slattery@CJSlattery
I’m getting some pushback from software guys, telling me I am undervaluing the full feature set of pieces of software. Let me say that the full feature set used to its maximum potential is a useless metric. If 70% of your customers use 10% of your software’s capabilities, an AI replacement only needs to replace that bottom 10%, likely the most simple 10%, for it to cut your revenue by 70%. Almost nobody uses your product fully. And the more complex the product the smaller the percentage of users who take full advantage. I don’t need to replace everything triplewhale does to take a massive bite out of it. I just need to replace the 5 things that 50% of the user base uses it for. Now this is something SaaS can work to prevent by more aggressively getting customers to use more of the tool. But the fact remains, the actual value derived from average users is far below the theoretical maximum.
English
almost a month into using pi.dev
- still the most predictable output
- claude code unusable
- codex struggling
the biggest lever for running agentic engineering at scale is predictability.
Stuart Chaney@stuartchaney
made the switch to pi this week the predictability of the output has been incredible so far it’s extremely hard to scale agentic engineering at a real company without owning the system end to end Great intro video for anyone interested
English
Stuart Chaney retweetledi
Introducing HQ: the AI operating system for your company.
HQ lets your team share knowledge, skills, workflows and API access across the organization, so everyone works from the same foundation.
Team AI is still messy, and hard.
Individuals work in silos. There is no consistent knowledge base. Teams end up with 10 versions of the same skill, all performing at different levels. One person becomes 10x more effective, while everyone else is left trying to catch up.
HQ makes this simple.
Built on top of Codex and Claude Code, HQ sits directly inside your existing workflow. It helps manage your company's context layer, share skills, standardize workflows and control access, so people only see what they need to see.
We built HQ internally to solve a real problem. We use it every day. It has been incredible for enabling our power users, but just as importantly, it has helped the least technical members of our team become incredibly effective AI users.
HQ turns one person's breakthrough into everyone's baseline.
It is a powerful, elegant tool, and we are incredibly excited to share it with the public.
Book a demo below.
English
100% agree
We are still very far away from taking humans out of the loop in the art of shipping code to production.
Dillon Mulroy@dillon_mulroy
i still read every line of code btw, often many times over
English
@couuor overrides some commonly used tools on your machine to compress the output (and strip the noise out) before it hits the context window.
English
Claude's billing + usage limits are extremely unintuitive, so I dug in over the weekend to find ways to reduce costs:
First, the cost I was accruing was on top of standard seat usage that I had on our team plan.
(1) Since we're on the team plan there's no Max option, but they have a 'Premium' seat which is similar to Max and 6.25x the usage limit
(2) Usage is allotted in 5 hour windows, so I began scheduling more tasks to utilize tokens in non-work hours.
(3) I had Claude rewrite some of my most used Skills so that it delegates tasks to subagents using more affordable models
(4) I updated claude.md in my main projects to be more concise, using a mix of Karpathy's 4 Rules and that Caveman approach that went viral a couple weeks ago.
I still hit overages over the weekend but was <$15 vs the $100+ I had spent on Friday alone and what sparked the initial question.
Connor MacDonald@couuor
I’m not developing software with Claude, but use it heavily for notes, strategy docs, etc—all the knowledge work I do. What is a reasonable month of token costs? How much are different members of your team spending? I spent ~$200 in April. Looking for more reference points.
English
33% reduction using /goal after an hour
still running and code changes look great so far...
Stuart Chaney@stuartchaney
/goal in codex is awesome reducing the duration of your test suite is a great place to start
English
/goal in codex is awesome
reducing the duration of your test suite is a great place to start
Felipe Coury 🦀@fcoury
/goal also lands in Codex CLI 0.128.0. Our take on the Ralph loop: keep a goal alive across turns. Don't stop until it's achieved. Built by my co-worker and OpenAI mentor Eric Traut, aka the Pyright guy. One of the GOATs I get to work with daily.
English
I fixed a lot of context window bloat with this prompt.
Less tokens = better output
Paste this into codex/claude code/pi
"LLMs get dumb after 150K tokens
Your goal is to analyze the contents of my initial context window and provide actionable steps to reduce the initial context without sacrificing quality.
Focus on low effort high impact changes. Quality over quantity"
English
ideal example of where agentic coding shines
feature: adding sparkline charts (like Shopify)
non-trivial update that would have taken a least a full day - now gets shipped to prod in under 45 mins.
builds svg components based on the timeseries data
note:
every codebase update must pass all of our guardrails/tests before it gets manually reviewed by a human.
English
Stuart Chaney retweetledi
Agentic coding is a huge boon for software developers that want to get far more done, great for IT people to build vastly more custom systems internally, great for domain experts that want to automate workflows or wire systems together, and absolutely fantastic for anyone curious to learn how to start coding.
What it’s less great for is casually building complex software that you have to maintain on an ongoing basis and take on all the risk for. Upgrades, maintenance, keeping up to date with latest security issues, and so on, are taxes most knowledge workers aren’t familiar with or prepared for.
Net net: we’re going to get 100X more software and vastly more software developers in the future. But that’s different from *everyone* rolling their own.
English
@29nextdevin it’s very very good at following direction, much like 5.4 - but just hasn’t clicked with my codebase yet. Working through it though
English
@stuartchaney How do you feel about codex? GPT 5.5 seems to have that intuitive vibe back
English
April will be the lowest month for code commits in over a year
strong correlation with the opus fall off
currently rebuilding systems as the models evolve and change - healthy to always assume its a skill issue.
Nov - March:
Kill the guardrails / complex workflows. The models just figure it out.
April:
Bring back the guardrails. The models take things very literally now.
onwards
English
1. follow @mattpocockuk
2. use his /grill-me skill
by far my most used skill this year for any medium/large codebase change.
Matt Pocock@mattpocockuk
Nearly 23K stars for a collection of markdown files I wrote I guess they must be pretty good I want to invest more time in this repo. So, folks who starred it, what can I do to make these skills more obvious to you? - A docs site for the skills? - Send them to plugin marketplaces? Help me help you github.com/mattpocock/ski…
English
@jacob_posel gpt-5.5 medium
opus-4.7 also works much better than inside of claude code - but will use gpt with my sub instead.
English
made the switch to pi this week
the predictability of the output has been incredible so far
it’s extremely hard to scale agentic engineering at a real company without owning the system end to end
Great intro video for anyone interested
English