Software effort estimation is fundamental for planning and controlling software projects and critical for their success. The practice has shown that realistic goals setting, accurate estimates, and good planning and control are essential activities for project success, even for agile endeavors. This paper focuses on how Colombian developers perform effort estimates within agile projects. For this purpose, we conducted an exploratory survey study that involved 60 respondents. The main findings are the following: (1) agile practitioners prefer non-algorithmic estimation techniques, mainly those based on Expert Judgement. (2) Most of the respondents perceive that their estimates have a medium accuracy level; however, in 70\% of the cases no formal analysis of the error magnitude is carried out. (3) The determining effort predictors/cost drivers are characteristics of the project team (size, experience, and skills) and attributes of the software to be built (complexity, type, and domain). (4) The use of datasets for estimation is not common, the proprietary ones predominate, and are used for productivity comparisons within the company. (5) Most of the results of related studies are comparable with ours; however, there are significant differences in terms of the roles involved in the estimation process and the methods used.