Here’s a new episode of The Reddy Rundown, crafted so you don’t have to frantically follow everything in the AI news space wondering what you are missing as a an exec in 2025 trying to keep up

I am Shawn Reddy, CEO and Founder at AI Advantage Agency. I think like a systems first marketer who now runs a company. My take this week is simple. The winners will not be the loudest brands. The winners will be the teams that wire new models into repeatable systems that shrink time to value and prove it in the funnel.

GPT 5 is lining up for August. Plan for a reasoning upgrade, not a cosmetic one

Multiple outlets report OpenAI is preparing GPT 5 for an August launch, with Sam Altman hinting at a jump in practical reasoning during a long form interview this week. Expect a unification of language and o series style reasoning that reduces model switching across tasks. Translation for marketers. Less orchestration glue, faster creative iteration loops, stronger agents for quoting, triage, and research. I am planning evaluation harnesses around real funnel tasks, not chat benchmarks. Budget a controlled pilot on day one, and set a kill switch if conversion or response quality dips during the first week.

Comet moves from shiny browser to actual assistant for meetings

Perplexity’s Comet connects to Gmail and Calendar to brief you on the day, mine your inbox, and draft follow ups. The useful part for marketing is not the summary. It is the pre meeting research pack and the auto generated questions tied to the companies on your calendar. That makes a one person growth team feel like three. Use it to walk into partner and prospect calls with tailored proof points and fresh news.

Guardrails are now a conversion lever, not just a safety checkbox

Fiddler’s new enterprise benchmarks compare cost, latency, and accuracy across jailbreak resistance, toxicity control, and faithfulness. Treat guardrails like landing page QA for your agents. Cleaner answers and fewer hallucinations reduce the number of human reviews and keep reply times low, which lifts conversion on support and sales handoff flows. Pick policies per stage of your funnel instead of one blanket setting.

What people actually do with AI at work keeps shifting the task mix

Microsoft researchers analyzed 200 thousand Bing Copilot conversations and found the highest exposure in computer science, office support, sales, and media roles, with AI mostly providing information, writing, teaching, and advising. That supports a strategy where SDRs and content teams get agent support first, with clear rules for when to trust or escalate. Structure your enablement around draft then verify workflows rather than fully autonomous creation.

A wild card from bio shows the speed curve we should copy

DTU researchers published an AI platform that designs minibinder proteins in weeks, guiding T cells to target cancer markers. The marketing relevance is not medicine. It is the pace. When discovery cycles compress from years to weeks, your creative testing and offer testing should mirror that cadence. Shorten your loop from idea to shipped to measured.

Infra note worth watching for latency and cost

SambaNova’s SN40L platform is now in AWS Marketplace and optimized for top open models like Llama and DeepSeek. If your chat or quoting load is spiky, this matters. Shorter response times and the ability to switch models quickly can reduce drop offs in high traffic widgets and keep your agent stack responsive during campaigns. It also makes procurement simpler for enterprise buyers who prefer AWS channels.

What I am changing in my own playbook

1. I am moving evaluation to task level outcomes, like booked meeting rate and time to first reply, not generic accuracy scores.

2. I am splitting agents by outcome. One for quote generation and one for service triage, each with its own guardrail profile.

3. I am adding pre meeting briefs to every partner and enterprise call, sourced automatically from Comet style connectors plus our own CRM context.

4. I am setting a fixed two week window after any model swap where we run A and B traffic to catch silent regressions before they hit revenue.

Tools and how to use them this week

• Perplexity Comet use it to auto brief every meeting and prep three tailored questions tied to the company on your calendar also have it draft the follow up with the exact next step you want.

• Fiddler Guardrails Benchmarks choose guardrail settings per funnel stage strict for public chat and lighter for sales assisted flows measure impact on reply time and escalation rate.

• SambaNova Cloud on AWS test an open model agent behind your website quote widget and compare latency and abandonment against your current setup during a live campaign.

• Microsoft study use the occupation exposure findings to prioritize enablement start with SDRs and content ops where AI already aligns with the tasks people delegate.

Looking for a community of like-minded individuals who are interested in AI and Entrepreneurship? Join our free community here to get started: The AI Advantage Community Thank you for reading! -Shawn