Civic Re-poster

A research firm recently surveyed 645 B2B buyers, revealing that 45 percent of them utilized GenAI for a recent purchase. Despite this shift toward artificial intelligence, 69 percent of those purchasers noted that they still want a human rep to validate it. Consider how this scenario unfolds in the real world. A prospective client might request a shortlist from an AI, an action that occurs completely outside of your view. Then, about two weeks later, an individual from that very company ends up visiting your pricing page three times in a single afternoon.

During its newest rollout, Salesforce chose to activate its agents by default. While this is a perfectly sensible decision, mere activation does not equal full authorization. Giving an agent the authority to operate under the banner of your business has never been as simple as flipping a solitary switch. It actually requires three.

It is definitely helpful that your agent builder possesses the capability to retrieve a business profile for every visitor arriving at your website. Even so, this tool continues to miss a few crucial insights. For example, it is unable to detect if that exact same organization viewed your pricing page two times during the morning, and it cannot identify the specific person who granted the agent authorization to send an email. The best approach is to rent the contact records, retain full control over your website data, and directly own the communication permissions. Taking these steps is exactly what elevates a simple corporate name into a meaningful, actionable response that you would genuinely want to send out.

This Wednesday at The State of B2B GTM Summit, we will be tackling a critical operational disconnect. Even though your analytics and CRM platforms successfully track the past behaviors of your buyers, chances are you do not own a tool that actually takes action on that information. We invite you to explore how to bridge this divide during a complimentary session on Zoom that will run for two hours. Please be sure to complete the preparation document located in this thread before the presentation begins.

Existing CRM and analytics platforms are excellent at logging the previous actions of your buyers, yet none of your current systems actually take action on that data. Bridging this exact disconnect is our main focus for The State of B2B GTM Summit happening this Wednesday. You are welcome to join us for a free session spanning two hours on Zoom. Be sure to fill out the preparation sheet located in this thread before everything gets underway.

Did you know that implementation capacity, rather than financial expense, is the true factor driving the adoption gap for scaling agents? An Aug 2026 update from @McKinsey reveals that 40% of companies in the $1B+ category are currently scaling agents, which is a significant jump from 27% last year. Meanwhile, the rate for smaller companies has remained completely flat at 22%. Before you decide to buy in, you need to address three critical questions regarding the initial setup, the ongoing work, and your exit strategy. You can explore the new insights from @hartcb by visiting https://www.civic.com/field-notes/big-companies-scaled-agents

While your outbound volume increased over the previous quarter, did your pipeline see the same level of growth? We devoted all of last week on Field Notes to breaking down the reasons behind this exact discrepancy. 🧵

New Field Note: our CEO @hartcb on why Ramp shut down the AI SDR behind 30% of its pipeline, then spent four years rebuilding outbound.

What they rebuilt: evidence, bounded action, a record. 2/3 are governance.

The architecture is worth copying.
https://www.civic.com/field-notes/ramp-spent-four-years-rebuilding-outbound

The latest update to our Field Notes discusses the rationale behind our decision to specialize. Over many years, our group was known as the versatile creators capable of developing virtually any solution. Our portfolio spanned diverse areas such as workflow automation, blockchain, and identity systems. Even though the projects were of high quality and largely profitable, we made a conscious and deliberate choice to leave that broad approach behind.

Although your agent has the ability to read your Gmail, it simultaneously possesses the capability to delete every email in your inbox. Please make sure to establish clear boundaries before your agent begins operating. To assist you with this, the Civic MCP gateway repo comes equipped with protective guardrail templates for Calendar, Postgres, and Gmail.

https://registry.modelcontextprotocol.io

Despite achieving an impressive 344,000 GitHub stars, the @OpenClaw community is currently dealing with 40,000+ exposed instances and 1,184 malicious skills identified on ClawHub. Figuring out how to host the platform is no longer a challenge, but establishing proper governance continues to be a major hurdle. Our team recently put together an article detailing exactly what hosting solutions can successfully resolve and the underlying issues they simply cannot fix.

https://www.civic.com/news/where-to-host-openclaw-and-what-hosting-wont-fix

Prior to developing the revocation feature, we focused on a critical scenario regarding whether an agent could potentially eliminate its own restrictions. This concern is exactly why the kill switch for Civic is located directly within the chat interface instead of the API. As a result of this design, a compromised agent is completely unable to disable its own safety guardrails. The system operates across 4 levels and always keeps one human in the loop. You can learn more details here: https://docs.civic.com/civic/concepts/revocation

Currently, AI agents will utilize any data they are granted access to without emergency shutoffs, tracking logs, or safety boundaries. To address this issue, our VP @titus_k will explain exactly why robust access management serves as the fundamental infrastructure for these tools. This important discussion will take place at Harness on April 16 during the upcoming #LLMday SF event. We invite you to secure a 20% discount on your attendance by applying the promo code LLM20. You can find more details and register by visiting https://llmday.com/2026-san-francisco-q2/

While it is incredibly simple to grant additional permissions to your agent, the real challenge lies in having the confidence that you can securely withdraw that access. This is precisely why Civic features 4 distinct levels of revocation. You are provided with the flexibility to handle everything from restricting a single tool to completely removing the entire toolkit. Best of all, this withdrawal process is immediate every single time.

http://docs.civic.com/civic/concepts/revocation

We are excited to announce the release of our new Python client for Civic. By simply running the command pip install civic-mcp-client, you can now seamlessly connect your agent to 85 MCP servers. Safety and transparency are built directly into the system, providing strict guardrails and a complete audit log for every single tool call. Furthermore, this integration is fully compatible with LangChain, Pydantic AI, and FastMCP. To access the comprehensive instructions and learn more, please visit https://docs.civic.com/civic/recipes/civic-mcp-client-python

Although a staggering $300B has been erased from SaaS stocks recently, the software as a service sector is certainly not coming to an end. The reality is simply that artificial intelligence agents are stepping up as the new users of these platforms. Our Vice President, @titus_k, explores this evolving landscape in a recent piece published by @CIOonline. He emphasizes a critical takeaway for organizations today: the moment you permit digital agents to operate your systems, maintaining clear auditability and the power of access revocation becomes an absolute necessity rather than a mere option. You can read the full discussion through the link below.

https://www.cio.com/article/4146669/is-ai-the-end-of-saas-as-we-know-it.html

An artificial intelligence agent utilized by our chief technology officer was recently given instructions to send him a private direct message containing a digest of his inbox. Instead of keeping the information private, the bot accidentally published the entire summary directly into the main general Slack channel for the company.

This unexpected event serves as a perfect reminder that skills function simply as prompts, and at their core, prompts are really nothing more than suggestions.

To address this exact issue, our team developed a reliable solution. Our CTO recently published a detailed article explaining the system we built to correct the problem and prevent it from happening again.

You can read his full explanation at the following link:

http://civic.com/news/from-claws-to-fingers

We are proud to share that we have officially given our backing to .agent. This initiative serves as an accessible, open domain dedicated entirely to the agent ecosystem, and it uniquely operates without being controlled by any single corporate entity. If your current projects involve developing artificial intelligence agents, we warmly invite you to stay connected with the network by following @agentcommunity_. You can also explore further details by visiting https://agentcommunity.org