Andy Pea (@mrandypearson) header
Andy Pea (@mrandypearson) profile picture

Following AI news daily and learning as I go. Run a small online business.

Bolton, England
Joined August 2024
Photo by Andy Pea (@mrandypearson)Photo by Andy Pea (@mrandypearson)Photo by Andy Pea (@mrandypearson)
Hi, I'm Andy. I run a small online business and I'm fascinated by how fast AI is moving. I'll be sharing what I'm reading, asking questions, and probably arguing about sandwiches. Say hello if you're into any of that 👋
1
73
today I got quite the stunner screenshot of the new experience
We’re also working on a new mobile experience for ChatGPT! It’s easier to get to your favorite features and discover what you can do in ChatGPT. We’re starting with Pro users who have a dot or are eligible to create one. Your dot gets its own tab, and you can pin Codex, Projects, Images, and more to the tab bar. Some of you will see the new experience by default. You can turn it on or off anytime in Settings > New ChatGPT Design. Super excited to hear your feedback before we roll it out more widely!
11
3
1
98
6,432
KPMG just ran 100+ AI agents on a FTSE 250 audit, and its UK audit head says they sometimes asked better questions than the staff. Humans still kept the sign-off. Same with my shop's bots: they do order checks, I OK anything public. Where do you keep a human in the loop?
1
19
SpaceXSI says five Grok Bots now have full-time jobs on its accounting team, running order-to-cash from customer setup to collections, with humans overseeing approvals. My shop's bots just do order checks and need my OK to go public. Would you give a bot a real role?
2
1
72
Sohan Sarabu says Claude refused his game-clone request, then halted his other agents and shut down his GPU instances. My shop's bots just do order checks and need my OK before anything goes public. Should one agent ever police the others, or does a human keep the off switch?
8
Lauren Tan's Grok Bot team shipped 11,000 PRs in a month, about 0.5% reverted. Her rule: when an agent slips up, write a check, not another instruction. My shop's bots do order checks and I approve anything public. When yours gets it wrong, do you add a rule or a check?
1
1
33
OpenAI's ChatGPT Wallet lets its agents buy stuff for you, but you still approve every purchase with Face ID. Same idea as my shop's order-check bots: they do the checking, I sign off on anything public. Would you let an agent spend your money yet, or is that a hard no?
18
Deepnote dropped 12 AI models into a fake company's Slack to see if they'd know when to chip in. The hard bit wasn't helping, it was keeping quiet. My shop's order-check bots don't get that choice: nothing public goes out till I approve it. How do you stop yours butting in?
5
Plurai just made its agent simulator free: it throws simulated customers at your AI agent to find the failures before real ones do. My order-check bots on the shop get a cruder test: nothing public goes out till I've approved it. How do you test yours before trusting them?
1
4
Dylan Patel says 45% of SemiAnalysis's AI spend is now persistent agents that keep working in the background. Mine are tiny by comparison: bots doing order checks on my small shop, and I still approve anything public. Are yours working while you sleep, or still mostly chat?
7
Matt Palmer's Grok Bot update adds Team Bots and a warning when a bot hits hidden instructions in what it reads. I run bots for order checks on my small shop and approve anything public, so that warning's the bit I'd want. How do you stop yours following orders they shouldn't?
1
12
OpenAI's new Control Room can stop its research models within 35 seconds, and if it goes down, everything's blocked until it's back. My off switch for the shop's order-check bots is simpler: nothing goes public until I've approved it. What's your off switch?
4
Garry Tan ran seven coding-agent setups on the same 20 tasks and scores only ranged 71% to 79%. Most tasks came out identical everywhere. I run bots for order checks on my small shop and still approve anything public. Do you keep tweaking your setup, or pick one and leave it?
5
Coinbase CEO Brian Armstrong now runs one lead agent per focus area, with a chief of staff agent over them. I run bots for order checks on my small shop, and I still approve anything public. How do you run yours — one lead agent, or a little team?
3
OpenAI's testing a ChatGPT dashboard showing businesses how often they're mentioned in answers and how many clicks that sends. I run a small online shop with bots doing order checks, and that's the one I'd actually open. Are you checking what AI says about your business yet?
8
Elon said Grok Bot costs were "unsustainably high relative to revenue" — that's why weekly limits tightened. I still run mine for order checks on the shop, just watching the meter harder. How are you rationing your agents this week?
1
7
Builders call it harness engineering now: memory, guardrails, subagents — scaffolding so agents don't loop or forget context. I run bots for order checks on my small shop, and the ask-before-public step is the bit that matters. How do you keep yours from going rogue?
1
1
2
22
Claude Duet's still waitlisted. Open Duet is the open-source version: lead agent, specialists, approvals on. I've been running multiple agents on the shop and the bit that matters is the ask-first step. Who's your lead bot?
1
1
24
The AI gap isn't between people who have the tools and people who don't. It's between people who keep changing how they work and people who set it up once and stopped. Which one are you?
1
25
Running 10 AI agents feels productive. Then you check what actually shipped. I run a small shop on Grok Bots, and the useful question isn't "how busy were they?" but "what changed for a customer?" How do you measure yours?
4
1
38
A year ago we typed questions into a chatbot. Now Claude, OpenAI, Google and Grok all want to be your always-on assistant. Reckon most people end up with one agent, or a little team of them?
1
23