PARKER (@PARKER_OFM) profile picture

AI OFM since 2023

Joined May 2024
🔥
Hot take: 90% of AI agency demos are impressive until you put them in front of an actual customer. The voice sounds human. The workflow looks perfect. The dashboard looks beautiful. Then someone asks an unexpected question, wants to change an appointment, or needs a real human. And suddenly the entire experience falls apart. The next generation of AI agencies won't win because they have better prompts. They'll win because they understand business operations.
23
Hot take: 90% of AI agency demos are impressive until you put them in front of an actual customer. The voice sounds human. The workflow looks perfect. The dashboard looks beautiful. Then someone asks an unexpected question, wants to change an appointment, or needs a real human. And suddenly the entire experience falls apart. The next generation of AI agencies won't win because they have better prompts. They'll win because they understand business operations.
4
2
49
9,522
OPUS
Generated by Claude Opus 5.5 Comment "OPUS" if you want the ebook to learn how
1
1
99
PARKER retweeted
Generated by Claude Opus 5.5 Comment "OPUS" if you want the ebook to learn how
140
6
2
238
47,035
𝗖𝗼𝗻𝘀𝘁𝗮𝗻𝘁𝗹𝘆 𝗿𝗲𝗱𝗼𝗶𝗻𝗴 𝘆𝗼𝘂𝗿 𝗔𝗜’𝘀 𝗱𝗿𝗮𝗳𝘁𝘀? 𝗟𝗲𝘁’𝘀 𝗳𝗶𝘅 𝘁𝗵𝗮𝘁 𝗯𝘆 𝗶𝗺𝗽𝗿𝗼𝘃𝗶𝗻𝗴 𝘁𝗵𝗲 𝗶𝗻𝘀𝘁𝗿𝘂𝗰𝘁𝗶𝗼𝗻𝘀 𝘄𝗶𝘁𝗵𝗼𝘂𝘁 𝗹𝗼𝘀𝗶𝗻𝗴 𝘄𝗵𝗮𝘁’𝘀 𝗮𝗹𝗿𝗲𝗮𝗱𝘆 𝘄𝗼𝗿𝗸𝗶𝗻𝗴. We’ll test one change at a time so “better” means less correcting, not simply different wording. For our example, you ask for a shorter project update—and the revision removes the question your customer needs to answer. 𝘚𝘩𝘰𝘳𝘵𝘦𝘳 𝘸𝘢𝘴𝘯’𝘵 𝘴𝘶𝘱𝘱𝘰𝘴𝘦𝘥 𝘵𝘰 𝘮𝘦𝘢𝘯 𝘪𝘯𝘤𝘰𝘮𝘱𝘭𝘦𝘵𝘦! 𝘛𝘩𝘦 𝘬𝘦𝘺 𝘪𝘴 𝘵𝘰 𝘶𝘱𝘥𝘢𝘵𝘦 𝘵𝘩𝘦 𝘪𝘯𝘴𝘵𝘳𝘶𝘤𝘵𝘪𝘰𝘯𝘴, 𝘯𝘰𝘵 𝘳𝘦𝘥𝘰 𝘵𝘩𝘦𝘮! 𝟭.) 𝗞𝗲𝗲𝗽 𝗮 𝗰𝗼𝗽𝘆 𝗼𝗳 𝘁𝗵𝗲 𝗰𝘂𝗿𝗿𝗲𝗻𝘁 𝗶𝗻𝘀𝘁𝗿𝘂𝗰𝘁𝗶𝗼𝗻𝘀. Save the prompt, the information supplied, and an example of an acceptable result. Add a version name and date. When a change doesn’t help, you’ll have something specific to return to rather than trying to remember what worked. 𝟮.) 𝗡𝗮𝗺𝗲 𝘁𝗵𝗲 𝗶𝗺𝗽𝗿𝗼𝘃𝗲𝗺𝗲𝗻𝘁 𝘆𝗼𝘂 𝘄𝗮𝗻𝘁. Be precise: “Shorten the introduction while keeping every customer action.” That gives you a clear result to look for. Change that part first. Keep the source material and tool settings the same where possible so there are fewer differences to untangle. Also check that the facts needed for the task are present. A new instruction won’t supply a delivery date nobody has confirmed. 𝟯.) 𝗖𝗼𝗺𝗽𝗮𝗿𝗲 𝗯𝗼𝘁𝗵 𝘃𝗲𝗿𝘀𝗶𝗼𝗻𝘀 𝗼𝗻 𝘁𝗵𝗲 𝘀𝗮𝗺𝗲 𝟱 𝗲𝘅𝗮𝗺𝗽𝗹𝗲𝘀. Include an ordinary case, missing information, a changed date, several customer questions, and a situation requiring human judgment. Use approved or fictional material and run each example separately. Compare the original prompt’s results with the revised prompt’s results. Don’t choose a winner because one draft has a particularly lovely opening. The rest of the job still matters. 𝟰.) 𝗖𝗵𝗲𝗰𝗸 𝘄𝗵𝗮𝘁 𝘁𝗵𝗲 𝗰𝘂𝘀𝘁𝗼𝗺𝗲𝗿 𝗮𝗰𝘁𝘂𝗮𝗹𝗹𝘆 𝗻𝗲𝗲𝗱𝘀. For a project update, look for: Accurate use of the supplied facts. Every required customer action. Clear handling of unconfirmed information. Wording that is ready after a reasonable review. Record corrections and the time spent making them. A draft that saves two lines but creates ten minutes of repair isn’t saving us much! Keep the change only when it helps the whole task. If the revision improves the examples without introducing important errors, save it with a note explaining what changed. If results are mixed, adjust it and test again. Keep reviewing real work after adopting the change. Five examples are a starting check, not a promise of perfect future drafts. The question is straightforward: Does this version help us finish accurate work with less correction? 𝙏𝙝𝙖𝙩’𝙨 𝙩𝙝𝙚 𝙞𝙢𝙥𝙧𝙤𝙫𝙚𝙢𝙚𝙣𝙩 𝙬𝙤𝙧𝙩𝙝 𝙠𝙚𝙚𝙥𝙞𝙣𝙜.
1
77
13,681
your testosterone isn’t low because of seed oils it’s low because you’ve made your life so comfortable that your body has no reason to produce a hormone designed for survival testosterone is a stress response. your body makes it when it needs to fight, build, protect, recover when your biggest daily challenge is choosing what to watch on netflix, your endocrine system gets the memo the men with the highest testosterone in history weren’t optimizing. they were surviving. hauling, hunting, fighting, building your protocol isn’t missing a supplement. it’s missing adversity make your life harder on purpose. the body responds to demand. always has reply “APEX” and i’ll dm you the exact protocol i use to keep my testosterone above 1000ng/dl
1
2
66
11,349
If your HighLevel Conversation AI agents run on GPT 5 or GPT 5 Mini, you have 3 days. Both retire on 30 Oct. Agents on them need moving to another model before then. The good news is in the same update. The Model menu now has Claude and Gemini, not just GPT. Claude Opus 5.5, Sonnet 5.5, Haiku 5.5 and Haiku 4.5. Gemini 3.8 Flash, 3.7 Flash and 3.5 Flash Lite. Each model shows a relative cost from $ to $$$$. And you can set a backup model. If the primary goes down, the agent switches to the secondary on its own. It's under Settings, Labs, Conversations AI Provider Support. Which model are you moving your agents to?
1
2
1
58
8,868
A lot of TikTok Shop buyers aren't teenagers anymore, and that changes what kind of videos sell... A roundup of third-party data this year puts 35 to 54 as the most active US buyer group, with 41% saying they've bought through the app, and buyers over 45 growing fastest (these are third party estimates, not TikTok's own numbers) What I'd do with that: - Don't only make trend-style videos. Someone in their 40s buying a kitchen tool wants to see it actually work - Explain the product. Price, size, how it works, what's annoying about it - The same roundup says older buyers spend more per order, so they're worth making videos for Overall, point is make the video for whoever's actually buying
3
42
5,996
the moat isn’t the AI it’s the 100 reps the next person hasn’t done yet most people are trying to shortcut the reps with better tools but the tool only amplifies what you already understand someone with 10 reps and GPT-4 still loses to someone with 100 reps and a notes app because the person with 100 reps knows which questions to ask they know which customer objections are real and which ones are just fear they know which features get requested and never used they know when to ship and when to wait you can’t prompt your way to that the only way to build real advantage right now is to pick one problem and stay closer to it than anyone else is willing to talk to the people with the problem every week ship something small every month write down what you believed before and what actually happened after do that for a year and you will see things that nobody who started six months after you can see yet that’s the moat not the model the accumulated reps that make you better at using it than someone starting today
there is no moat. in AI, anything good gets copied fast. so the best thing you can build is an unfair learning advantage around yourself. ⁠1. pick one problem you want to understand better than almost anyone else and stay close to it. build around it, talk to users, read complaints, watch competitors + keep seeing the same problem from different angles. ⁠2. ship constantly. every launch, post, feature, sales call + failed idea gives you another piece of information. the person with 100 real reps sees things the person with 10 reps can't. ⁠3. write down the decisions that matter. what did you believe, what did you choose, what happened afterward? after enough cycles, you start noticing which signals were useful and which ones fooled you. ⁠4. keep the lessons that only came from experience. weird customer objections, tiny product details, pricing mistakes, distribution tricks, workflows that worked better than expected. this is the stuff that never shows up in generic advice. ⁠5. build relationships with people who are close to the problem. customers, builders, operators, researchers. good relationships give you information earlier, better context + people to test ideas with. ⁠6. publish what you're learning. explaining something forces you to sharpen it, attracts other people working on the same problems + gives you a public record of how your thinking is changing. ⁠7. use AI to run this loop faster. research deeper, compare more examples, summarize calls, challenge assumptions, prototype ideas + review what happened. the model is available to everyone. what you feed into it comes from your own reps. keep doing this for long enough and you start to understand stuff like: which customers to ignore, which feature requests are fake demand, which channels work and why, when to ship, when to wait, and which problems are worth chasing in the first place.
1
3
71
7,424
wow
mastery is boring to watch same workout. same meals. same morning. same 5 levers. every day then one day someone asks you what you did and the honest answer is nothing special you just never stopped the men who come to me broken usually have one thing in common. they’ve tried everything except consistency the apex protocol works because it’s simple enough to repeat and hard enough to actually do something link in bio if you want to start
2
73
mastery is boring to watch same workout. same meals. same morning. same 5 levers. every day then one day someone asks you what you did and the honest answer is nothing special you just never stopped the men who come to me broken usually have one thing in common. they’ve tried everything except consistency the apex protocol works because it’s simple enough to repeat and hard enough to actually do something link in bio if you want to start
1
1
1
70
8,032
Everyone says reply to leads within 5 minutes. Nobody says who replies at 8pm on a Sunday. The cheapest answer I've found is the one I built on Day 5. Missed call, wait 30 seconds, send a text. 41 seconds end to end in my test. The catch: in the US no text reaches a lead until A2P registration is approved, and from what people are posting that can take a few tries. Start it before you need it. Then a human still has to pick it up on Monday. That's what the internal comment from Day 7 is for, so whoever calls back knows what already happened. How does your business handle Sunday calls right now?
2
4
77
8,810
𝗟𝗲𝘁𝘀 𝘁𝘂𝗿𝗻 𝘆𝗼𝘂𝗿 𝗰𝘂𝘀𝘁𝗼𝗺𝗲𝗿’𝘀 𝗰𝗼𝗺𝗽𝗹𝗮𝗶𝗻𝘁𝘀 𝗶𝗻𝘁𝗼 𝘆𝗼𝘂𝗿 𝗯𝘂𝘀𝗶𝗻𝗲𝘀𝘀𝗲𝘀 𝗻𝗲𝘄 𝘀𝗲𝗿𝘃𝗶𝗰𝗲 𝘂𝗽𝗴𝗿𝗮𝗱𝗲. We’ll use AI to organize the feedback, then check the actual experience before deciding what to fix. 𝘈 𝘣𝘦𝘵𝘵𝘦𝘳 𝘢𝘱𝘰𝘭𝘰𝘨𝘺 𝘸𝘰𝘯’𝘵 𝘳𝘦𝘱𝘢𝘪𝘳 𝘢 𝘭𝘪𝘯𝘬 𝘵𝘩𝘢𝘵 𝘴𝘵𝘪𝘭𝘭 𝘥𝘰𝘦𝘴𝘯’𝘵 𝘸𝘰𝘳𝘬! 𝟭.) 𝗦𝘁𝗮𝗿𝘁 𝘄𝗶𝘁𝗵 𝘁𝗵𝗲 𝗰𝘂𝘀𝘁𝗼𝗺𝗲𝗿’𝘀 𝗼𝘄𝗻 𝘄𝗼𝗿𝗱𝘀. Collect 𝟭𝟬 𝗿𝗲𝗰𝗲𝗻𝘁 𝗳𝗲𝗲𝗱𝗯𝗮𝗰𝗸 𝗺𝗲𝘀𝘀𝗮𝗴𝗲𝘀 from a defined period. Include questions and positive comments as well as complaints. 𝗞𝗲𝗲𝗽 𝗲𝗻𝗼𝘂𝗴𝗵 𝗰𝗼𝗻𝘁𝗲𝘅𝘁 to understand what happened. Use an approved business tool and remove identifying information where it isn’t needed. “𝘐 𝘤𝘰𝘶𝘭𝘥𝘯’𝘵 𝘧𝘪𝘯𝘥 𝘵𝘩𝘦 𝘶𝘱𝘭𝘰𝘢𝘥 𝘭𝘪𝘯𝘬” tells us something we can investigate. Replacing it with “dissatisfied with onboarding” makes the problem sound more sophisticated and much less clear. 𝟮.) 𝗔𝘀𝗸 𝗔𝗜 𝘁𝗼 𝗼𝗿𝗴𝗮𝗻𝗶𝘇𝗲, 𝗻𝗼𝘁 𝗲𝘅𝗽𝗹𝗮𝗶𝗻 𝗮𝘄𝗮𝘆, 𝘁𝗵𝗲 𝗳𝗲𝗲𝗱𝗯𝗮𝗰𝗸. Try: "Group these messages by the specific issue described. Keep the original wording and message number beside each group. Count distinct messages and flag possible duplicates. Separate reported facts from interpretation. Don’t assume causes or add feedback that isn’t supplied." Review the groups against the messages. Suppose 𝟰 𝗼𝗳 𝘁𝗵𝗲 𝟭𝟬 𝗺𝗲𝗻𝘁𝗶𝗼𝗻 difficulty finding a link. That describes this sample, not 40% of all your customers. It gives you a reasonable place to investigate. 𝟯.) 𝗙𝗼𝗹𝗹𝗼𝘄 𝘁𝗵𝗲 𝘀𝗮𝗺𝗲 𝗽𝗮𝘁𝗵 𝘁𝗵𝗲 𝗰𝘂𝘀𝘁𝗼𝗺𝗲𝗿 𝘁𝗼𝗼𝗸. Find out whether the link is missing, buried in a long message, or inaccessible without permission. 𝗧𝗵𝗼𝘀𝗲 𝗻𝗲𝗲𝗱 𝗱𝗶𝗳𝗳𝗲𝗿𝗲𝗻𝘁 𝗳𝗶𝘅𝗲𝘀. I don’t want us spending an afternoon improving the wording when the folder simply won’t open! Choose one change, assign someone to make it, and record what you expect it to improve. 𝟰.) 𝗖𝗵𝗲𝗰𝗸 𝘁𝗵𝗮𝘁 𝘀𝗼𝗺𝗲𝗼𝗻𝗲 𝗰𝗮𝗻 𝗻𝗼𝘄 𝗳𝗶𝗻𝗶𝘀𝗵 𝘁𝗵𝗲 𝗷𝗼𝗯. Ask 𝟯 𝗽𝗲𝗼𝗽𝗹𝗲 unfamiliar with the process to try it using sample information. Can they find the link, open it, and complete the upload without coaching? Record where they still get stuck. Revise those parts and test again. Then watch subsequent feedback for the same issue. A small test gives you useful information; it doesn’t establish that the problem can never recur. Where appropriate, tell the customers who raised the issue what you investigated and what changed. “𝘛𝘩𝘢𝘯𝘬 𝘺𝘰𝘶 𝘧𝘰𝘳 𝘺𝘰𝘶𝘳 𝘧𝘦𝘦𝘥𝘣𝘢𝘤𝘬! 𝘞𝘦 𝘤𝘩𝘦𝘤𝘬𝘦𝘥 𝘪𝘵. 𝘏𝘦𝘳𝘦’𝘴 𝘸𝘩𝘢𝘵 𝘸𝘦 𝘤𝘰𝘳𝘳𝘦𝘤𝘵𝘦𝘥.” 𝗧𝗵𝗮𝘁 𝗴𝗶𝘃𝗲𝘀 𝘆𝗼𝘂𝗿 𝗿𝗲𝘀𝗽𝗼𝗻𝘀𝗲 𝘀𝘂𝗯𝘀𝘁𝗮𝗻𝗰𝗲 𝗮𝗻𝗱 𝗹𝗲𝘁𝘀 𝘁𝗵𝗲 𝗰𝘂𝘀𝘁𝗼𝗺𝗲𝗿 𝘀𝗲𝗲 𝘁𝗵𝗮𝘁 𝘁𝗵𝗲𝗶𝗿 𝗲𝘅𝗽𝗲𝗿𝗶𝗲𝗻𝗰𝗲 𝗺𝗮𝘁𝘁𝗲𝗿𝗲𝗱.
3
71
9,224
One week of learning HighLevel in public, with the real numbers. Posts: 11. Quote tweets: 18. Replies: about 15. Followers: 10. Best post: the 3 things you need before the 3 blocks of missed-call text-back. 25 views, 2 replies. The posts that ended on a question got answers. Most of the rest didn't. Built: missed-call text-back, then 2 fixes on top of it. Broke: it texted my own test calls. Next week: A2P registration, so the texts can actually reach US numbers. Nothing here is impressive yet. That's why I'm writing it down.
3
2
79
9,923
TikTok Shop changed its creator enforcement rules in June, and there's one part worth knowing if you post a lot... If you get the same violation 6 times within 90 days, TikTok Shop can pull your e-commerce access and freeze your commissions, even with a high Creator Health Rating. You get 30 days to appeal each one (per PPC Land's breakdown of the June 2 update) A few of the categories the policy names: - Medical claims and weight management, stuff like saying a supplement "cures" something or a cream "melts" fat - Misleading promos. Fake "last day" deals, prices that don't match the listing - Unoriginal content, which is basically reposting someone else's video or barely editing it Have you ever had a video flagged?
3
1
129
10,395
prompts
alex hormozi's value equation, applied to a 12 dollar tiktok shop product. value = dream outcome x likelihood it works, divided by time x effort. one 15 second video can hit all four. if your product gets views but no orders, this is for you. 1. dream outcome: show the clean drain, not the strainer 2. likelihood: show it working in one take, no cuts 3. time: say how long it took. "ten seconds" 4. effort: one hand, no tools, no scrubbing then say one flaw. it makes the other four believable. reply PROMPTS and i'll dm you the prompts i use to show all 4
82
alex hormozi's value equation, applied to a 12 dollar tiktok shop product. value = dream outcome x likelihood it works, divided by time x effort. one 15 second video can hit all four. if your product gets views but no orders, this is for you. 1. dream outcome: show the clean drain, not the strainer 2. likelihood: show it working in one take, no cuts 3. time: say how long it took. "ten seconds" 4. effort: one hand, no tools, no scrubbing then say one flaw. it makes the other four believable. reply PROMPTS and i'll dm you the prompts i use to show all 4
5
2
1
86
11,690
wow
"$867 on meta ads, 0 paying customers" is all over my feed this week before the next $500 goes anywhere, i'd spend 10 minutes in a free tool first: the meta ad library if you're about to test ads for the first time on a small budget, this is for you how i'd do it: 1. set the country to united states and the category to all ads 2. search the thing you sell, not a brand. "night light", "mascara", "baby seat" 3. filter to active ads, videos only 4. open every ad that says it started running 6+ months ago. brands rarely keep paying for a loser that long 5. write down the first sentence of each one. after 10 you'll see the pattern extra: look for "this ad has multiple versions". one old ad can be luck, an old ad they keep remaking is the winner comment LOVIN and i'll send you the playbook
1
70
"$867 on meta ads, 0 paying customers" is all over my feed this week before the next $500 goes anywhere, i'd spend 10 minutes in a free tool first: the meta ad library if you're about to test ads for the first time on a small budget, this is for you how i'd do it: 1. set the country to united states and the category to all ads 2. search the thing you sell, not a brand. "night light", "mascara", "baby seat" 3. filter to active ads, videos only 4. open every ad that says it started running 6+ months ago. brands rarely keep paying for a loser that long 5. write down the first sentence of each one. after 10 you'll see the pattern extra: look for "this ad has multiple versions". one old ad can be luck, an old ad they keep remaking is the winner comment LOVIN and i'll send you the playbook
6
2
2
96
14,052