Agentic AI trendsfor 2027.
Ten shifts that will define how companies put AI agents to work in customer service, sales and learning. It runs 55 pages with data from primary sources, verified quotes, charts and real screens, plus a 90-day plan to move beyond the pilot.



The ebook in four numbers.
- of organizations at least experiment with AI agents, but only 23% scale them in some area
- 62%
- Source: McKinsey, The state of AI in 2025
- of agentic AI projects are expected to be canceled by the end of 2027, according to Gartner
- 40%+
- Source: Gartner, Jun 25, 2025
- is how long it takes for the length of tasks agents complete with 50% success to double
- 7 months
- Source: METR, Mar 19, 2025
- of customers say companies using GenAI in customer service must offer access to a person
- 87%
- Source: Gartner, Aug 4, 2026
What changes with AI agents in 2027.
Each trend takes three pages in the ebook: what changes, the analysis with numbers and a quote, and what to do in practice, with a real screen from live operations.
01
Agents that work for hours and use screens
The length of tasks an agent completes on its own doubles roughly every seven months, and agents already operate browsers and systems without APIs. Clear processes get far more out of them.
02
Real-time AI voice calls
Speech-to-speech models accept interruptions and query systems during the call. Confirmations, collections and first-level support are the first use cases.
03
Agents that learn from operations
Persistent memory and learning from the team's corrections make the agent improve with use, with human review of what it comes to know.
04
Smaller, specialized models
Agents use more tokens than a chat. Routing between small and frontier models decides whether the agent fits the budget.
05
AI CRM fed by conversations
Records come from the conversation: the agent qualifies, creates the opportunity and sets the next step. Without unified customer data, the sales agent fails.
06
AI customer service with people in the loop
Agents resolve the routine end to end and hand the rest to the team with context. Customers accept the agent when they know they can reach a person.
07
Agentic commerce
Protocols such as AP2 and the Agentic Commerce Protocol give the agent a mandate to buy. The next customer may arrive as an agent comparing offers.
08
Agent teams and open protocols
An orchestrator distributes work among specialists, and MCP and A2A, under the Linux Foundation, connect agents to systems and to each other.
09
Agent identity and security
OWASP listed the ten risks of agentic applications. Every agent needs an identity, minimum permissions and a log of every action.
10
Regulation and governance
The European AI Act already requires telling people they are talking to an AI, and bill PL 2.338/2023 is moving through Brazil's Congress. Documented governance comes before the rule.
55 pages to take to the committee.
Prose and analysis
Each trend explained in full paragraphs, with the context that leads to it and the impact on customer service, sales and operations.
Charts and numbers
Bars, donuts, timelines and diagrams, always with the source and date at the foot of the page.
Verified quotes
Words from the people who research and build the technology, checked against the primary source, with the original English at the end of the ebook.
Real operation screens
Real captures of CRM, voice calls, knowledge base, operations and audit logs, to see each trend in practice.
90-day plan
A roadmap to move beyond the pilot: choose three processes, build with minimal tools and operate with the numbers on the table.
Ten questions for the committee
The list to take to the meeting where the agent is approved, from autonomy and permissions to decision logs.
Who wrote it.

Marlos Carmo
Co-founder of Tolky
More than 25 years of experience in digital. One of the 500 entrepreneurs of the Economy of Francesco and a finalist of the SXSW Innovation Awards.

Patrick Bonnereau
Co-founder of Tolky
Educator and product manager, physicist from UFMG and Google Innovator. Finalist of the SXSW Innovation Awards and awarded by Microsoft in Paris.
Main sources
01
Gartner · Jun 25, 2025
Gartner Predicts Over 40% of Agentic AI Projects Will Be Canceled by End of 202702
Gartner · Aug 26, 2025
Gartner Predicts 40% of Enterprise Apps Will Feature Task-Specific AI Agents by 202603
McKinsey · Nov 2025
The state of AI in 2025: Agents, innovation, and transformation04
METR · Mar 19, 2025
Measuring AI Ability to Complete Long Tasks05
Salesforce · Feb 03, 2026
State of Sales, 7th edition06
Anthropic · Sep 29, 2025
Managing context on the Claude Developer Platform07
Linux Foundation · Dec 09, 2025
Linux Foundation Announces the Formation of the Agentic AI Foundation08
OWASP · Dec 09, 2025
OWASP Top 10 for Agentic Applications for 202609
Comissão Europeia · Jul 27, 2026
AI Act Service Desk, Article 5010
Agência Senado · Dec 10, 2024
Senado aprova regulamentação da inteligência artificial; texto vai à Câmara
Agentic AI trends for 2027
Agentic AI is AI that receives a goal and carries out the steps to reach it: it plans, uses tools and systems, keeps context and checks the result, within limits set by the company.
The ebook brings together ten: agents that work for hours and use screens, AI voice calls, agents that learn from operations, specialized models, AI CRM, AI customer service with people in the loop, agentic commerce, agent teams and open protocols, agent security and regulation.
Yes. The PDF is free and downloads directly from this page, with no sign-up.
According to Gartner, because of rising costs, unclear business value and inadequate risk controls. The ebook shows how to choose processes with a clear return and measure from day one.
The PDF is in Brazilian Portuguese, and this page summarizes it in English. Quotes published in English appear in a free translation in the PDF, with the original text on the final pages.
Yes, in the opening, on an advertising page in the middle and in the closing, plus discreet boxes in the trends the platform already delivers, such as AI CRM, AI customer service, AI voice and learning from the team.
Put an agent into productionin 2027.
Tolky brings together AI CRM, AI customer service, AI voice calls and agents that learn from your team, on the same platform. Let's talk?
