Darryl Hicks

Passionate serial entrepreneur, angel investor, active advisor to startups, hopeless payments geek, currently scaling VC backed #fintech Revaly #Bitcoin fan

[Darryl Hicks @darryl_hq]

While I’d love to see broader adoption of modern payment rails, Chamath and the All-In squad miss an important insight (discussion around stable coins at the 10:00min mark) into what drives the majority of online transactions into VISA/MC/AMEX rails.

Spoiler alert: It has almost nothing to do with merchant choice, and incumbent bias doesn’t even crack the top three reasons. Let’s break it down in this thread:

[The All-In Podcast @theallinpod]

E208 is LIVE!

@DOGE kills its first bill, zuck vs @openai, and google's ai comeback with bestie aaron @levie

(0:00) bestie intros

(3:07) the besties welcome @box ceo aaron levie!

(5:40) thoughts on @DavidSacks new role and what areas he can positively impact quickly in crypto and ai

(18:42) @DOGE's first casualty and its paradigm-shifting potential

(48:25) conspiracy corner: what are these drones over new jersey!?

(57:36) zuck joins team @elonmusk vs openai, ai's competitive landscape, downstream impact on business software

(1:27:29) google's ai turnaround, future of ai physics and video models

[Darryl Hicks @darryl_hq]

@tonysimons_ Here’s my Git Repo. Lots of human and machine readable files in there, just point your Hermes agent at the repo and say “what from this would be an upgrade for our setup” and let it rip

https://t.co/8jSShjWmFL

[Darryl Hicks @darryl_hq]

Built an assistant to my CoS, staff accountant with some deep skills, coder / engineer (love using Grok 4.5 for red team’ing my GPT 5.6 builds), research / analyst for the Office of the CEO, deep knowledge repo on all IP in my biz. Sovereign AI systems FTW! Shared some of my fav builds on my git repo, DM me if you’re interested in seeing what I’ve built. Love sharing and learning from this community!

[Darryl Hicks @darryl_hq]

Mac Mini 32 GB
used to run open claw, migrated to Hermes for stability
GPT 5.5 reasoning = high for most work, GPT 5.4 OR 5.4 mini for simpler tasks. Opus 4.8 for copy, GPT 5.5 via API for deeper builds where the 1m context window helps (OAuth limited to 272k)
Primary use = assistant to my CoS, and analyst in the Office of the CEO
5 sub agents / profiles
Researcher
Call transcriber
Coder / engineer (by far the most robust)
Staff accountant
Copywriter (trained on my voice, and anti-AI slop git repos)

Burnt 3.8bn tokens last 30 days. Mostly on custom builds for dashboards to help the ELT and me stay on top of board level initiatives, as well as deep processing of signals from recorded calls (Fathom and Plaud), Teams, Asana, iMessage with VIPs, and chats, all fed into LanceDB indexed memory vault and Obsidian. Game changing tech.

[Darryl Hicks @darryl_hq]

@tonysimons_ Reporting for duty! I maintain a public git repo with all my favorite builds on Hermes and have a few active WhatsApp chats for founders who like to build.

[Darryl Hicks @darryl_hq]

Also super important to setup your agent so that views anything it sees as information only, it should only accept prompts from white listed sources like your telegram / discord chat. Cuts out 99% of your prompt injection risks.

I had a buddy who got prompt injected by a PDF attachment to an email with white txt on a white background. He was having all email attachments converted to .md files by his bot and got hit. Brutal.

[Darryl Hicks @darryl_hq]

Deep LLM processing of all my calls, slack, teams, iMessage, asana, and select emails i bcc my agent on to build a deep indexed repository of everything going on in my business, in the context of my 2026 goals, so anyone on my ELT can get a deep picture of where we’re at, and i have a dashboard with red / green lights letting me k kw where we’re on / off track on key initiatives.

[Darryl Hicks @darryl_hq]

It’s tied to the broader theme of “where do I let the LLM work it’s magic” vs “where do I impose hard gates”. Over-prescribing means you end up with milquetoast automation. If you under-specify you end up with chaos. For my vault I use gates. File cannot be stored unless it passes the gate logic, temp working files obviously excluded.

[Darryl Hicks @darryl_hq]

I just switched to Hermes. The cutover was painful as I have 5 sub agents and over 80 cron jobs, many of which needed to be migrated to Hermes native tooling, but overall insanely more stable. Much slower, but gets things almost entirely right on first try. And no crashes. Even did an update, no bricking and no major issues, only some minor cleanup post update. I think it’s gonna be way better.

[Darryl Hicks @darryl_hq]

Thx for sharing! I just nuked my system by upgrading to 5.12 and have spent all morning mopping up the mess after reverting to 5.7. The documented telegram bug in 5.12 but me hard, they fixed it with a merged PR in 5.14 beta but when I upgraded to that one out of desperation it had a whole hot mess that pegged my cpu at 100% and crippled the whole system. Will try your tool! Thx for sharing.

[Darryl Hicks @darryl_hq]

@openclaw For the love of all that is good and holy, can we please get a release that doesn’t brick my system? All cron jobs failing, telegram 10x slower, have to force “visible send” to get any response in topics in telegram. And now I have to revert, again. What a waste of time.

[Darryl Hicks @darryl_hq]

13/ We're actively deploying those rails and have seen great success in our POC. The ones waiting for networks to figure it out will watch approvals drop.

[Darryl Hicks @darryl_hq]

Your AI assistant just bought you groceries. The bank declined it. Not because it was fraudulent. Because the bank has no idea who, or what, just tried to spend your money. A thread 🧵

[Darryl Hicks @darryl_hq]

12/ Agentic commerce is coming whether payments infrastructure is ready or not. The merchants and issuers who start building trust into the rails now, with bi-directional data sharing and auditable signals, will capture this wave.