{"id":385,"date":"2025-07-11T23:47:58","date_gmt":"2025-07-11T23:47:58","guid":{"rendered":"https:\/\/zoltrunakiver.com\/?p=385"},"modified":"2025-07-21T14:30:09","modified_gmt":"2025-07-21T14:30:09","slug":"5-different-paths-to-launch-ai-agents-with-examples","status":"publish","type":"post","link":"https:\/\/zoltrunakiver.com\/index.php\/2025\/07\/11\/5-different-paths-to-launch-ai-agents-with-examples\/","title":{"rendered":"5 different paths to launch AI agents (with examples)"},"content":{"rendered":"
Hello and welcome to The GTM Newsletter by GTMnow <\/strong>\u2013 read by 50,000+ to scale their companies and careers. GTMnow shares insight around the go-to-market strategies responsible for explosive company growth. GTMnow highlights the strategies, along with the stories from the top 1% of GTM executives, VCs, and founders behind these strategies and companies.<\/em><\/p>\n Everyone\u2019s talking about AI agents. Some teams are quietly building. Others are still wondering where to start. And a few are already running huge parts of their orgs on agents.<\/p>\n How do you actually launch one that gets the job done?<\/p>\n This edition walks through five different paths to launch an AI agent<\/strong>.<\/p>\n Each path includes:<\/p>\n It also outlines how each path maps to a broader AI adoption curve, inspired by Google\u2019s maturity model. All together, the paths mapped against Google\u2019s model looks like this:<\/p>\n Here\u2019s a breakdown of each, with real examples.<\/p>\n If you\u2019re just getting started with AI agents, this is the lowest-lift and fastest path to real leverage. No engineers required.<\/p>\n This path is best for automating glue work: outbound, follow-ups, lead research, CRM updates. The kind of stuff early-stage teams often put on a SDR or virtual assistant.<\/p>\n Use this when you\u2019re: <\/strong><\/p>\n Common mistakes to be aware of: <\/strong><\/p>\n Prompt quality and targeting. Quality can dip fast if you don\u2019t tune your prompts or tighten your ICP filters. And without CRM tracking, you won\u2019t know what\u2019s working.<\/p>\n What it looks like in practice:<\/strong><\/p>\n This can take every form that you can essentially imagine. Here\u2019s one example:<\/p>\n If your team already has some structure (regular pipeline reviews, onboarding processes, forecast prep, etc.) agents can help you do the same work with fewer cycles.<\/p>\n This path works by wrapping a lightweight AI layer around something you already do consistently. It doesn\u2019t require you to invent a new system, just automate the parts that drain time or block velocity.<\/p>\n Use this when you\u2019re: <\/strong><\/p>\n Common mistakes to be aware of:<\/strong><\/p>\n Assuming the agent can do the full workflow out of the gate. Start with narrow tasks (summarize \u2192 post \u2192 suggest), not complex decision-making. Human-in-the-loop QA is still key at this stage.<\/p>\n What it looks like in practice:<\/strong><\/p>\n One example:<\/p>\n Another example:<\/p>\n Agent build with Lindy.ai<\/a><\/em><\/p>\n<\/div>\n<\/div>\n<\/figure>\n<\/div>\n This is our GTMnow podcast guest research AI agent. Every time we add a new guest to a spreadsheet, the AI agent jumps into action researching, creating a document, and attaching that document to the guest.<\/p>\n We still spend a minimum of three hours doing additional human research, including listening to past episodes, but it makes the process far more efficient.<\/p>\n If you\u2019re juggling things like hiring, pipeline recaps, creating investor or board updates and it all lives in your head (or in GSheets, Notion, etc.) a custom GPT copilot can be your silent, reliable teammate.<\/p>\n This path works best for repetitive communication tasks where the format is known, the content is semi-structured, and you want to stay inside your existing tools.<\/p>\n Use this if you\u2019re:<\/strong><\/p>\n Common mistakes to be aware of:<\/strong><\/p>\n GPTs are great at structure, but not nuance. If your tone matters (to investors, recruits, or customers), build a few reference samples into the prompt or system message. And always layer in human QA before hitting publish.<\/p>\n What it looks like in practice:<\/strong><\/p>\n Here\u2019s one example:<\/p>\n When your customers are asking the same 20 questions every day or your team is buried in call notes and follow-ups, embedded agents can step in and take the first pass.<\/p>\n These agents live inside the tools you already use, such as Intercom, Gong, HelpScout, Zendesk. They respond to support questions, summarize calls, or even prep CRM updates \u2013 all without a human ever touching the task.<\/p>\n Use this if you\u2019re:<\/strong><\/p>\n Common mistakes to be aware of: What it looks like in practice:<\/strong><\/p>\n Here\u2019s one example:<\/p>\n This is the most powerful (and the most complex) way to launch an AI agent. You\u2019ll need technical horsepower, but the upside is huge: full control, deep automation, and the ability to build something no off-the-shelf tool can match.<\/p>\n These custom agents can drive multi-step onboarding flows, real-time sales coaching, multi-agent task routing \u2013 truly whatever your stack and imagination allow.<\/p>\n Use this if you\u2019re:<\/strong><\/p>\n Common mistakes to be aware of:<\/strong><\/p>\n Agent fragility and maintenance debt. These systems are powerful, but they break if the stack changes or the prompts aren\u2019t updated regularly. Build with versioning, fallbacks, and human escalation paths baked in.<\/p>\n What it looks like in practice:<\/strong><\/p>\n Here\u2019s one example:<\/p>\n Jordan Crawford<\/a> built an AI agent workflow that queries a Snowflake database of 172 million permits to find the top 3 most relevant ones for each prospect \u2014 data that\u2019s independently valuable to them.<\/p>\n It costs just ~$0.30 per query, and it\u2019s like having a personalized data science team for every lead. The flow works by:<\/p>\n You can read or listen about this agent here<\/a>.<\/p>\n These five agent paths can serve as a selection guide, but more importantly they represent stages in a broader AI maturity journey.<\/p>\n Google maps AI adoption<\/a> across three (plus) levels:<\/p>\n This same progression applies to GTM teams adopting AI agents.<\/p>\n To bring this to life even more, here\u2019s how a common GTM task evolves as your AI maturity increases: outbound follow-up.<\/p>\n Each stage adds more automation, context, and scale. This gives you more leverage.<\/p>\n Start where you are. Just automate one thing, then continue layering on from there. The leverage will compound.<\/p>\n \u201cThe worst thing teams can do is overthink it. Don\u2019t spend 6 weeks on a spreadsheet. Just start. Pick a use case. Build an agent. Test it.\u201d<\/em><\/p>\n \u2013 Ray Smith (VP of AI Agents, Microsoft)<\/em><\/p>\n<\/blockquote>\n<\/div>\n For more helpful resources on AI agents: <\/strong><\/p>\n
\n<\/div>\n\n
<\/div>\n<\/figure>\n<\/div>\n
PATH 1. The no-code starter<\/h2>\n
\n
<\/div>\n<\/figure>\n<\/div>\n
PATH 2. Wrap an agent around an existing workflow<\/h2>\n
\n
\n
<\/p>\n
PATH 3. Build a custom GPT copilot<\/h2>\n
\n
<\/div>\n<\/figure>\n<\/div>\n
PATH 4. Embed agents in your GTM stack<\/h2>\n
\n
\n<\/strong>Voice and fallback. These agents are<\/em> your brand in moments that matter. Make sure the tone feels human, and that fallback paths (like escalating to a rep) are clear and fast. A bad support experience will impact your brand.<\/p>\n<\/div>\n<\/figure>\n<\/div>\n
PATH 5. Build from scratch<\/h2>\n
\n
<\/div>\n<\/figure>\n<\/div>\n
\n
\n
<\/div>\n<\/figure>\n<\/div>\n
<\/div>\n<\/figure>\n<\/div>\n
\n
\n
\n<\/div>\n