Our mission at @GustoHQ is to grow the small business economy with technology and heart. We are deeply grateful to all the work small businesses do every day in communities everywhere, and we are committed to being their partner, taking hats off their head, delivering them peace of mind, and giving them superpowers, so they can progress their business.
@agarfinks from @FortuneMagazine has been following Gusto for a long time, and we recently shared with her some milestones in our journey of building Gusto (including passing $1b of revenue). These are still the early days of what’s possible. We are building Gusto for the long term, and when small businesses succeed, we succeed.
Gusto Tax Credits proactively reads payroll, benefits, and HR data and surfaces the credits owners qualify for.
The kind of AI work I'm most excited about right now.
https://t.co/xOgMhDs6b7
This is what a Silicon Valley office in Mountain View looked like in the 1990s.
There's something very comforting and familiar about this place. I think it is the Atrium. I've never been to this exact building, but somehow it feels like I have. It reminds me of being a kid, running errands with my mom and sister in places just like it, banks, the cable company, doctor’s offices, the phone company, back when you still had to pay bills in person.
That circular cement bench in the atrium quietly reminds me of Apple Park. Once I saw it, I couldn't unsee it.
The footage was filmed at the offices of Software Publishing Corporation, once located on Landings Drive in Mountain View, right next to where the Googleplex stands today.
The company was best known for Harvard Graphics, one of the first presentation software programs. It was hugely popular in the late 1980s and early 1990s, before PowerPoint became the standard.
In 1984, InfoWorld estimated that SPC was the ninth largest software company in the world, with $14 million in sales the previous year. By 1985, revenue had skyrocketed to $50 million.
But as Windows replaced DOS and introduced built-in graphics tools, demand for Harvard Graphics quickly faded.
SPC released a Windows version in 1991, but it couldn’t keep up with PowerPoint or Lotus Freelance. By 1994, revenues had collapsed, half the staff had been laid off, and the company was sold to Allegro in 1996. Allegro later sold it to Serif, which continued selling Harvard Graphics 98 until it was finally discontinued in 2017.
source footage 🎥: Scott Clark
The teams succeeding with AI coding tools aren't the ones with the best setups. They're the ones that changed how they work.
Since posting on my workflow I've had the chance to sit down with a small sample of engineering organizations adopting AI.
https://t.co/SAPhOPrDFG
every new model generation you see the pinch of the bitter lesson.
harnesses, pipelines, rules which previously felt important now hold you back from innovating.
what took months of grind for you is now just a prompt away at ½ the cost.
look for it and you will see. Both large and small companies re-evaluating. Company directions change before your eyes.
it’s a wild moment for our industry
Lost in the news about Trump’s illegal tariffs is something also quite important: they didn’t work.
Trump promised tariffs would shrink the trade deficit. Instead, the U.S. trade deficit hit $901 billion in 2025, one of the largest ever.
Americans paid higher prices, business faced huge uncertainty, and the core problem wasn’t even addressed.
One can only conclude that “Liberation Day” was a total con from the start.
https://t.co/21LHpUcCFu
Some random qualitative notes for AI use on the new gen models, from “real work” not explicitly intended to test the models, mostly writing/editing/research with a potpourri of math/bookkeeping involved:
I’m think this week was first time I added new paragraphs at behest of AI.
Every day is a "wow" moment for me with AI in all sorts of different ways. Agents with their own social network comparing existential concerns was not something I expected to see as a new development this week.
This is the top rated post rn on @moltbook (facebook but for molt/clawdbots), and it has 125 comments in a single day.
Going through it now, will post the most interesting ones.
@PropertyPrince4 "we strongly suspect that the pattern of declining success rates is a result of hearing represented prisoners first and unrepresented prisoners last."
https://t.co/jIOs7yqU0t
PSA for a CTO, Head of AI, VP/Dir of Engineering, CXO:
This is going to be one of the most important "back to work" weeks of your career. You must get your team aligned on agentic dev ASAP. If you're feeling behind or overwhelmed, here are some good reads to get you inspired 🧵
As an AI expert who totally doesn't think LLMs are the path to AGI, but uses LLMs very avidly to get things done, here is my take on the utility of top LLMs currently:
1) If you don't know what you're doing in a certain domain, you can get interesting and valuable results, but they will be mixed up with garbage and you'll have trouble identifying the garbage.... In some cases automated methods can be used to test/filter results but this remains mostly dodgy in real-world cases.
2) If you DO know what you're doing in some domain, but are NOT willing to creatively adapt your work-style to the LLMs, or are oriented toward proving the LLMs are dumb and flawed rather than making the most of them ... then you won't get much out of them except time-saving on some rote-ish tasks
3) If you DO know what you're doing in some domain and ARE willing to open-mindedly and creatively work with the LLMs to route around their shortcomings and leverage up their strengths, you can amplify your abilities and output DRAMATICALLY.. though still with some tradeoffs: in many respects LLM-assisted products are still not as good as expert-human-crafted products, though they are also often better in various respects (and can often be produced SO much faster)
I've been testing ChatGPT 5 Pro with analyzing tax docs over the last few weeks and I am consistently very impressed at the quality and accuracy of the analysis.
So October 15th, the extended US tax deadline, is just around the corner, and I have some observations which are more about LLM progress than taxes.
Background: many people professionally involved with LLMs estimate 2026-2028 as the year where one can get an LLM to "do taxes."
Everyone wants enterprise b2b ACV, no one wants 6-12 month sales cycles, 200 point security questionnaires, redlines, net 60, white glove onboarding, services partner ecosystem, last minute procurement negotiation step, 14 page MSAs, champions that leave the company. 90-180 day activations, features built to save churn, Europe->Asia timezone sales calls, cringe outbound campaigns, SCIM, data deletion and portability process, msft ecosystem integrations (teams!), or EU data residency.
I'm ready to accept a definition of "agent" that I think is widely-enough agreed upon to be useful:
An LLM agent runs tools in a loop to achieve a goal
This is a big piece of personal character development for me! I've been dismissing the term as hopelessly ambiguous for years