- Everyday AI
- Posts
- Agents, LLMs, or Algorithms? A Playbook for Choosing AI
Agents, LLMs, or Algorithms? A Playbook for Choosing AI
ChatGPT mobile app hits $2B, Google AI Overviews' 25% drop in publisher traffic, U.S. Gov. might take stake in Intel and more!
Everyday AI
August 15, 2025
š Subscribe Here | š£ Hire Us To Speak | š¤ Partner with Us | š¤ Grow with GenAI
Outsmart The Future
Today in Everyday AI
6 minute read
š Daily Podcast Episode: Choosing the right AI isn't simple with algorithms, LLMs, and agents vying for attention. Discover the nuances as we break down agentic AI, data's role, and risk management in AI projects. Give it a listen.
šµļøāāļø Fresh Finds: Sam Altman says weāre in an AI bubble, Metaās AI app still has flaws months later and NVIDIA unveils multilingual speech AI. Read on for Fresh Finds.
š Byte Sized Daily AI News: ChatGPT mobile app hits $2B global spending, Google AI Overviews linked to 25% drop in publisher traffic and U.S. Gov. might take a stake in Intel. For that and more, read on for Byte Sized News.
š§ Learn & Leveraging AI: Why spend hours trying to figure out the right AI workflow? Hereās how you can easily decide when to use what AI and how. Keep reading for that!
ā©ļø Donāt miss out: Did you miss our last newsletter? We talked about ChatGPT considering including ads, NSF and NVIDIA funding a scientific AI push, consumer groups calling out Grokās NSFW image tool and more. Check it here!
Agents, LLMs, or Algorithms? A Playbook for Choosing AI š¤
Confused by AI jargon and unsure which tools actually move the needle for your business?
We break down the real differences between traditional algorithms, large language models (LLMs), and agents ā including agentic AI ā and give practical guidance leaders can use now.
Also on the pod today:
ā¢ Importance of Data in AI Training š
ā¢ Risk Factors in Agentic AI Projects šØ
ā¢ Innovation through AI Experimentationš”
Itāll be worth your 33 minutes:
Listen on our site:
Click to listen
Subscribe and listen on your favorite podcast platform
Listen on:
Hereās our favorite AI finds from across the web:
New AI Tool Spotlight ā Aha is an AI-powered influencer marketing team, Twos is an AI-powered list and note-taking app, and the Shorty app helps you create viral videos with AI.
OpenAI ā While economists believe the stock market is in an AI bubble, Sam Altman has agreed and also believes AI is in a bubble.
NVIDIA ā NVIDIA has released an open data set for multilingual speech AI.
Meta ā Metaās AI app still has persistent flaws month after its launch.
Read This ā Check out this recent interview with Sam Altman as he discusses GPT-5 and more.
AI in Media ā This news station aired AI-generated images of woman soldiers.
1. ChatGPTās Mobile App Hits $2B in Consumer Spending š¤
ChatGPTās iOS and Android apps have reached $2 billion in global consumer spending since their May 2023 launch, pulling in $1.35 billion so far in 2025 and averaging roughly $193 million per month ā vastly outstripping rivals like Grok, Claude, and Copilot.
The data shows ChatGPT not only leads in revenue but also installs (about 690 million), with the U.S. accounting for 38% of spending and India the largest source of downloads, underscoring a broad consumer preference for mobile-first AI access.
2. Google AI Overviews Linked to 25% Drop in Publisher Referral Traffic šļø
According to Digital Content Next and the Professional Publishers Association, publishers are seeing year-over-year Google Search referral dropsāmedian declines of about 10% overall and up to 14% for non-news sitesālinked to Googleās new AI Overviews and rising zeroāclick behavior.
The data, collected MayāJune 2025 from DCN members and PPA submissions, shows steady CTR collapses (10ā25% in examples) even where rankings and impressions remain stable, suggesting AI summaries and Discover citations are replacing clicks. The timing is critical: with DOJ antitrust remedies and EU regulator pressure on the table, a court order to separate Googleās AI crawler from its search index could restore publishersā ability to opt out without vanishing from search results.
3. The U.S. Government Weighs Buying a Piece of Intel š
The U.S. government is in talks to take a direct stake in Intel to accelerate domestic chip manufacturing, including its delayed Ohio fab ā a move triggered after White House concerns about CEO LipāBu Tanās alleged China ties. The talks follow a tense August 11 meeting between Tan and the Trump administration and come days after President Trump publicly urged Tan to resign, underscoring heightened political scrutiny of the semiconductor supply chain.
If the government takes an ownership role, it could speed factory funding and regulatory support but also politicize corporate decisionsāraising questions for employees, suppliers, and startups that rely on Intel chips.
4. Anthropic Tightens Claude Rules to Block CBRN and Cyber Abuse āļøļø
Anthropic updated Claudeās usage policy to explicitly ban assistance in developing high-yield explosives and biological, nuclear, chemical, and radiological (CBRN) weapons, and added strict rules against using the model to discover or exploit computer and network vulnerabilities.
The move follows the rollout of āAI Safety Level 3ā protections with Claude Opus 4, aimed at making jailbreaks harder and curbing agentic features like Computer Use and Claude Code that could scale misuse. The changes also relax a blanket ban on political content to target only deceptive or disruptive campaign uses, and clarify that āhigh-riskā requirements ap
5. Cohere Raises $500M, Valuation Jumps to $6.8B š
Cohere said Thursday it closed an oversubscribed $500 million round, lifting its valuation to $6.8 billion ā up from $5.5 billion a year ago ā signaling renewed investor appetite for enterprise-focused LLMs. The startup is doubling down on a āsecurity-firstā pitch for businesses and has added senior hires like Joelle Pineau as chief AI officer and CFO Francois Chadwick to sharpen its enterprise play.
Strategic partners and backers including Radical Ventures, Inovia, Nvidia, AMD Ventures and Salesforce Ventures underscore momentum, while Oracleās participation this round remains unclear
6. Chinaās Power Edge Risks Widening AI Gap ā”
A new report from Tech Buzz China and interviews cited by Fortune say Chinaās deliberate overbuild of power ā with reserve margins often double need and rapid capacity additions ā leaves it free to scale AI data centers while the U.S. faces grid stress and permitting delays. According to Fortune, Goldman Sachs and Deloitte warn U.S. AI expansion is hitting a hard bottleneck as grids strain, pushing some firms to build private plants and stoking higher household bills.
The practical impact: companies and workers trying to grow AI businesses in the U.S. may face slower deployment, higher costs, and tougher site choices, while Chinese players can more quickly spin up capacity and iterate models.
š¦¾How You Can Leverage:
Michael Abramov runs a data labeling company with 480 employees.
Every single day, his calendar shows a blocked 30-minute slot.
Not for strategy meetings. Not for investor calls.
For playing with AI tools like a curious kid testing new toys.
This isnāt productivity theater. Michaelās company literally prepares datasets that train AI models for other businesses.
He watches AI startups get āblade cutā by Big Tech weekly. His survival depends on knowing which tools actually work versus which ones just look impressive on LinkedIn demos.
The uncomfortable truth about choosing AI? Most companies are making terrible decisions because theyāre chasing buzzwords instead of solving real problems.
We walked through the messy reality of picking between algorithms, large language models, and agents.
Michael shared his framework for avoiding the ācorporate bladeā thatās slicing through AI companies. Plus why being ādata-drivenā might be destroying your decision-making process.
Here are three insights that will change how you think about AI implementation.
1 ā Personal AI Experimentation Beats Team Workshops š§āš¬
Michael discovered something counterintuitive about AI adoption.
When CEOs experiment personally before delegating to teams, success rates skyrocket.
His approach eliminates what he calls the āmost people in the world problem.ā Employees make sweeping statements like āmost people are afraid of losing jobsā without any actual data. They extrapolate personal fears into universal truths.
Personal experimentation works differently.
You naturally gravitate toward your biggest daily pain points. Email chaos. Calendar nightmares. Slack overwhelm.
These become your organic testing ground.
Michael told us exactly how this plays out. You stop asking philosophical questions like āShould we use agents or LLMs?ā Instead, you ask tactical ones: āWhat fixes my email problem right now?ā
The magic happens when leaders share specific results, not general AI enthusiasm. Michaelās team sees concrete examples of time saved, not vague promises about transformation.
His company even runs weekly technical education sessions where employees show each other what actually worked. No formal presentations. Just real tools solving real problems.
Try This:
Block 30 minutes daily for two weeks. Pick your most frustrating daily task. Test three different approaches against it - a simple algorithm, an LLM tool, and an agent-based solution. Document actual time saved, not how impressed you felt.
Share specific results with your team using exact numbers and concrete outcomes.
2 ā The Corporate Blade Cuts Faster Than You Think š”ļø
The biggest threat to AI startups isnāt competition from other startups.
Itās getting sliced by giants who can replicate entire business models in weeks.
His example? Perplexity.
Multibillion-dollar valuation. Impressive growth metrics. But their entire value proposition is wrapping ChatGPT with better search capabilities.
OpenAI could recreate this in two weeks if they wanted.
This isnāt theoretical. Michael watches calendar assistants, AI therapists, and specialized agents get absorbed into major platforms monthly.
The pattern reveals something brutal. Most AI businesses are actually feature requests for existing platforms.
Companies building on foundation models without defensible moats? Theyāre essentially product managers for Big Tech.
When your entire value proposition becomes one menu option in ChatGPT, you donāt have a business.
Smart enterprises recognize this dynamic works both ways. Instead of betting on specialized AI vendors, theyāre developing internal capabilities that platforms canāt simply absorb.
Try This:
Audit your current AI vendor relationships this week. For each tool, ask one question: āCould this become a native ChatGPT feature within six months?ā If yes, start building internal alternatives immediately.
Focus your AI budget on tools that integrate with proprietary data or processes competitors cannot replicate.
3 ā Pseudo Data-Driven Thinking Kills AI Projects šµ
Michaelās team taught him something unexpected about being ādata-driven.ā
It often creates worse outcomes than making decisions with zero data.
After pushing his team to be more analytical, employees started presenting āproofā that proved nothing. Theyād correlate random metrics and claim causation.
Itās more dangerous than gut instinct because it appears scientific.
People relate things that are completely unrelated. They say things like āmost people in the world are afraidā without defining what āmostā means, which people, or where that data came from.
True data quality requires understanding what metrics actually predict AI success.
Most companies track the wrong things. Adoption rates instead of business impact. Chatbot engagement instead of resolution efficiency. Automation hours instead of customer satisfaction changes.
The companies winning with AI focus obsessively on outcome metrics that existed before AI arrived. Customer response times. Error rates. Revenue per employee.
If you canāt draw a clear line from your AI tool to improvement in pre-existing business metrics, youāre probably fooling yourself.
Try This:
Identify three AI tools your company uses right now. Stop measuring adoption metrics completely. Instead, track business outcomes that existed before AI - response times, error rates, or revenue per employee.
Set a 90-day deadline. If you cannot draw a direct line from the AI tool to measurable improvement in these pre-existing metrics, eliminate the project.
Reply